cotton_model.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow as tf\n",
    "from tensorflow.keras.models import Sequential\n",
    "from tensorflow.keras.layers import Dense, Conv2D, Flatten, Dropout, MaxPooling2D, ZeroPadding2D\n",
    "from tensorflow.keras.preprocessing.image import ImageDataGenerator\n",
    "from tensorflow.keras.utils import to_categorical\n",
    "from tensorflow.keras.models import Sequential\n",
    "from tensorflow.keras.layers import Dense, Conv2D, MaxPool2D , Flatten\n",
    "from tensorflow.keras.callbacks import ModelCheckpoint\n",
    "\n",
    "import skimage.io\n",
    "from selective_search import selective_search, box_filter\n",
    "from nms import nms\n",
    "\n",
    "import os\n",
    "from os import path\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.metrics import confusion_matrix, classification_report\n",
    "import cv2\n",
    "import json\n",
    "import random\n",
    "import tqdm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 253,
   "metadata": {},
   "outputs": [],
   "source": [
    "PATH = \"/mnt/d/personal/cotton/Annoted_boll_burst_2019_trial\""
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "PATH = \"/mnt/d/personal/cotton/samples/dset_rand/\""
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "train_dir = os.path.join(PATH, 'train')\n",
    "validation_dir = os.path.join(PATH, 'test')\n",
    "train_cats_dir = os.path.join(train_dir, 'positive')  # directory with our training cat pictures\n",
    "train_dogs_dir = os.path.join(train_dir, 'negative')  # directory with our training dog pictures\n",
    "validation_cats_dir = os.path.join(validation_dir, 'positive')  # directory with our validation cat pictures\n",
    "validation_dogs_dir = os.path.join(validation_dir, 'negative')  # directory with our validation dog pictures\n",
    "num_cats_tr = len(os.listdir(train_cats_dir))\n",
    "num_dogs_tr = len(os.listdir(train_dogs_dir))\n",
    "\n",
    "num_cats_val = len(os.listdir(validation_cats_dir))\n",
    "num_dogs_val = len(os.listdir(validation_dogs_dir))\n",
    "\n",
    "total_train = num_cats_tr + num_dogs_tr\n",
    "total_val = num_cats_val + num_dogs_val\n",
    "print('total training cat images:', num_cats_tr)\n",
    "print('total training dog images:', num_dogs_tr)\n",
    "\n",
    "print('total validation cat images:', num_cats_val)\n",
    "print('total validation dog images:', num_dogs_val)\n",
    "print(\"--\")\n",
    "print(\"Total training images:\", total_train)\n",
    "print(\"Total validation images:\", total_val)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "def read_all_images(pth):\n",
    "    imgs = []\n",
    "    for fl in os.listdir(pth):\n",
    "        imgs.append(cv2.imread(path.join(pth, fl)))\n",
    "    return np.array(imgs)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "X_1 = read_all_images(train_cats_dir)\n",
    "Y_1 = np.ones(X_1.shape[0])\n",
    "X_2 = read_all_images(train_dogs_dir)\n",
    "Y_2 = np.zeros(X_2.shape[0])\n",
    "X = np.concatenate((X_1, X_2))\n",
    "Y = to_categorical(np.concatenate((Y_1, Y_2)))\n",
    "X_1 = read_all_images(validation_cats_dir)\n",
    "Y_1 = np.ones(X_1.shape[0])\n",
    "X_2 = read_all_images(validation_dogs_dir)\n",
    "Y_2 = np.zeros(X_2.shape[0])\n",
    "XX = np.concatenate((X_1, X_2))\n",
    "YY = to_categorical(np.concatenate((Y_1, Y_2)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "batch_size = 128\n",
    "epochs = 30\n",
    "IMG_HEIGHT = 40\n",
    "IMG_WIDTH = 40"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 210,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "194380\n"
     ]
    }
   ],
   "source": [
    "train_image_generator = ImageDataGenerator(rescale=1./255,\n",
    "#                                           shear_range=0.2,\n",
    "#                                           vertical_flip=True,\n",
    "#                                           horizontal_flip=True,\n",
    "                                          width_shift_range=0.2,\n",
    "                                          height_shift_range=0.2\n",
    "                                          ) # Generator for our training data\n",
    "\n",
    "# train_image_generator = ImageDataGenerator(rescale=1./255)\n",
    "validation_image_generator = ImageDataGenerator(rescale=1./255) # Generator for our validation data\n",
    "# train_image_generator = ImageDataGenerator() # Generator for our training data\n",
    "# validation_image_generator = ImageDataGenerator() # Generator for our validation data\n",
    "\n",
    "train_data_gen = train_image_generator.flow(X, Y, shuffle=True, batch_size=batch_size)\n",
    "# val_data_gen = validation_image_generator.flow(XT, YT, batch_size=batch_size)\n",
    "total_train = len(X)\n",
    "print(len(X))\n",
    "# total_val = len(XT)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "train_data_gen = train_image_generator.flow_from_directory(batch_size=batch_size,\n",
    "                                                           directory=train_dir,\n",
    "                                                           shuffle=True,\n",
    "                                                           target_size=(IMG_HEIGHT, IMG_WIDTH),\n",
    "                                                           class_mode='categorical')\n",
    "val_data_gen = validation_image_generator.flow_from_directory(batch_size=batch_size,\n",
    "                                                              directory=validation_dir,\n",
    "                                                              target_size=(IMG_HEIGHT, IMG_WIDTH),\n",
    "                                                              class_mode='categorical')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "model = Sequential([\n",
    "    Conv2D(16, 3, padding='same', activation='relu', input_shape=(IMG_HEIGHT, IMG_WIDTH ,3)),\n",
    "    MaxPooling2D(),\n",
    "    Conv2D(32, 3, padding='same', activation='relu'),\n",
    "    MaxPooling2D(),\n",
    "    Conv2D(64, 3, padding='same', activation='relu'),\n",
    "    MaxPooling2D(),\n",
    "    Flatten(),\n",
    "    Dense(512, activation='relu'),\n",
    "    Dense(units=2, activation=\"softmax\")\n",
    "])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "model = Sequential([\n",
    "    Conv2D(16, 3, padding='same', activation='relu', \n",
    "           input_shape=(IMG_HEIGHT, IMG_WIDTH ,3)),\n",
    "    MaxPooling2D(),\n",
    "    Dropout(0.2),\n",
    "    Conv2D(32, 3, padding='same', activation='relu'),\n",
    "    MaxPooling2D(),\n",
    "    Conv2D(64, 3, padding='same', activation='relu'),\n",
    "    MaxPooling2D(),\n",
    "    Dropout(0.2),\n",
    "    Flatten(),\n",
    "    Dense(512, activation='relu'),\n",
    "    Dense(units=2, activation=\"softmax\")\n",
    "])\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# VGGNET\n",
    "model = Sequential()\n",
    "model.add(Conv2D(input_shape=(40,40,3),filters=64,kernel_size=(3,3),padding=\"same\", activation=\"relu\"))\n",
    "model.add(Conv2D(filters=64,kernel_size=(3,3),padding=\"same\", activation=\"relu\"))\n",
    "model.add(MaxPool2D(pool_size=(2,2),strides=(2,2)))\n",
    "model.add(Conv2D(filters=128, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\n",
    "model.add(Conv2D(filters=128, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\n",
    "model.add(MaxPool2D(pool_size=(2,2),strides=(2,2)))\n",
    "model.add(Conv2D(filters=256, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\n",
    "model.add(Conv2D(filters=256, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\n",
    "model.add(Conv2D(filters=256, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\n",
    "model.add(MaxPool2D(pool_size=(2,2),strides=(2,2)))\n",
    "model.add(Conv2D(filters=512, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\n",
    "model.add(Conv2D(filters=512, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\n",
    "model.add(Conv2D(filters=512, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\n",
    "model.add(MaxPool2D(pool_size=(2,2),strides=(2,2)))\n",
    "model.add(Conv2D(filters=512, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\n",
    "model.add(Conv2D(filters=512, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\n",
    "model.add(Conv2D(filters=512, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\n",
    "model.add(MaxPool2D(pool_size=(2,2),strides=(2,2)))\n",
    "model.add(Flatten())\n",
    "model.add(Dense(units=4096,activation=\"relu\"))\n",
    "model.add(Dense(units=4096,activation=\"relu\"))\n",
    "model.add(Dense(units=2, activation=\"softmax\"))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = Sequential()\n",
    "model.add(ZeroPadding2D((1,1),input_shape=(30, 30, 3)))\n",
    "model.add(Conv2D(64, 3,activation='relu'))\n",
    "model.add(ZeroPadding2D((1,1)))\n",
    "model.add(Conv2D(64, 3, activation='relu'))\n",
    "model.add(MaxPooling2D((2,2), strides=(2,2)))\n",
    "\n",
    "model.add(ZeroPadding2D((1,1)))\n",
    "model.add(Conv2D(128, 3, activation='relu'))\n",
    "model.add(ZeroPadding2D((1,1)))\n",
    "model.add(Conv2D(128, 3, activation='relu'))\n",
    "model.add(MaxPooling2D((2,2), strides=(2,2)))\n",
    "\n",
    "model.add(ZeroPadding2D((1,1)))\n",
    "model.add(Conv2D(256, 3, activation='relu'))\n",
    "model.add(ZeroPadding2D((1,1)))\n",
    "model.add(Conv2D(256, 3, activation='relu'))\n",
    "model.add(ZeroPadding2D((1,1)))\n",
    "model.add(Conv2D(256, 3, activation='relu'))\n",
    "model.add(MaxPooling2D((2,2), strides=(2,2)))\n",
    "\n",
    "model.add(ZeroPadding2D((1,1)))\n",
    "model.add(Conv2D(512, 3, activation='relu'))\n",
    "model.add(ZeroPadding2D((1,1)))\n",
    "model.add(Conv2D(512, 3, activation='relu'))\n",
    "model.add(ZeroPadding2D((1,1)))\n",
    "model.add(Conv2D(512, 3, activation='relu'))\n",
    "model.add(MaxPooling2D((2,2), strides=(2,2)))\n",
    "\n",
    "# model.add(ZeroPadding2D((1,1)))\n",
    "# model.add(Conv2D(512, 3, activation='relu'))\n",
    "# model.add(ZeroPadding2D((1,1)))\n",
    "# model.add(Conv2D(512, 3, activation='relu'))\n",
    "# model.add(ZeroPadding2D((1,1)))\n",
    "# model.add(Conv2D(512, 3, activation='relu'))\n",
    "# model.add(MaxPooling2D((2,2), strides=(2,2)))\n",
    "\n",
    "model.add(Flatten())\n",
    "model.add(Dense(4096, activation='relu'))\n",
    "model.add(Dropout(0.5))\n",
    "# model.add(Dense(4096, activation='relu'))\n",
    "# model.add(Dropout(0.5))\n",
    "model.add(Dense(2, activation='softmax'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 211,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow.keras.applications.mobilenet as mn\n",
    "model = mn.MobileNet(input_shape=(40, 40, 3), classes=2, weights=None)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 212,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(optimizer='adam',\n",
    "              loss=tf.keras.losses.categorical_crossentropy,\n",
    "              metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 113,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.load_weights(\"mn-09.hdf5\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "filepath=\"mn-{epoch:02d}.hdf5\"\n",
    "checkpoint = ModelCheckpoint(filepath,verbose=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 214,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:sample_weight modes were coerced from\n",
      "  ...\n",
      "    to  \n",
      "  ['...']\n",
      "Train for 1518 steps\n",
      "Epoch 1/10\n",
      "1517/1518 [============================>.] - ETA: 0s - loss: 0.0828 - accuracy: 0.9807\n",
      "Epoch 00001: saving model to mn-01.hdf5\n",
      "1518/1518 [==============================] - 1058s 697ms/step - loss: 0.0828 - accuracy: 0.9807\n",
      "Epoch 2/10\n",
      "1517/1518 [============================>.] - ETA: 0s - loss: 0.0771 - accuracy: 0.9821\n",
      "Epoch 00002: saving model to mn-02.hdf5\n",
      "1518/1518 [==============================] - 1022s 673ms/step - loss: 0.0771 - accuracy: 0.9821\n",
      "Epoch 3/10\n",
      "1517/1518 [============================>.] - ETA: 0s - loss: 0.0747 - accuracy: 0.9825\n",
      "Epoch 00003: saving model to mn-03.hdf5\n",
      "1518/1518 [==============================] - 1066s 703ms/step - loss: 0.0747 - accuracy: 0.9825\n",
      "Epoch 4/10\n",
      "1517/1518 [============================>.] - ETA: 0s - loss: 0.0702 - accuracy: 0.9836\n",
      "Epoch 00004: saving model to mn-04.hdf5\n",
      "1518/1518 [==============================] - 1036s 682ms/step - loss: 0.0701 - accuracy: 0.9836\n",
      "Epoch 5/10\n",
      "1517/1518 [============================>.] - ETA: 0s - loss: 0.0672 - accuracy: 0.9848\n",
      "Epoch 00005: saving model to mn-05.hdf5\n",
      "1518/1518 [==============================] - 993s 654ms/step - loss: 0.0672 - accuracy: 0.9848\n",
      "Epoch 6/10\n",
      "1517/1518 [============================>.] - ETA: 0s - loss: 0.0635 - accuracy: 0.9856\n",
      "Epoch 00006: saving model to mn-06.hdf5\n",
      "1518/1518 [==============================] - 1034s 681ms/step - loss: 0.0635 - accuracy: 0.9856\n",
      "Epoch 7/10\n",
      "1517/1518 [============================>.] - ETA: 0s - loss: 0.0624 - accuracy: 0.9858\n",
      "Epoch 00007: saving model to mn-07.hdf5\n",
      "1518/1518 [==============================] - 1041s 686ms/step - loss: 0.0624 - accuracy: 0.9858\n",
      "Epoch 8/10\n",
      "1517/1518 [============================>.] - ETA: 0s - loss: 0.0582 - accuracy: 0.9861\n",
      "Epoch 00008: saving model to mn-08.hdf5\n",
      "1518/1518 [==============================] - 1042s 687ms/step - loss: 0.0581 - accuracy: 0.9861\n",
      "Epoch 9/10\n",
      "1517/1518 [============================>.] - ETA: 0s - loss: 0.0552 - accuracy: 0.9872\n",
      "Epoch 00009: saving model to mn-09.hdf5\n",
      "1518/1518 [==============================] - 1043s 687ms/step - loss: 0.0552 - accuracy: 0.9872\n",
      "Epoch 10/10\n",
      "1517/1518 [============================>.] - ETA: 0s - loss: 0.0528 - accuracy: 0.9874\n",
      "Epoch 00010: saving model to mn-10.hdf5\n",
      "1518/1518 [==============================] - 1122s 739ms/step - loss: 0.0528 - accuracy: 0.9874\n"
     ]
    }
   ],
   "source": [
    "history = model.fit_generator(\n",
    "    train_data_gen,\n",
    "    steps_per_epoch=total_train // batch_size,\n",
    "#     steps_per_epoch = 2,\n",
    "#     epochs=epochs\n",
    "    epochs=10,\n",
    "    callbacks=[checkpoint],\n",
    "    class_weight={0:1, 1: 10}\n",
    "    \n",
    "#     validation_data=val_data_gen,\n",
    "#     validation_steps=total_val // batch_size\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 233,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.load_weights(\"mn-08.hdf5\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 208,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.save_weights(\"mn_jan_v2.h5\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "for layer in model.layers:\n",
    "    print(layer.name)\n",
    "    layer.trainable = True\n",
    "    print(layer.trainable)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "model.layers[-1].trainable = True\n",
    "model.layers[-2].trainable = True\n",
    "model.layers[-3].trainable = True\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "history = model.fit(\n",
    "    X, Y,\n",
    "    epochs=30,\n",
    "    validation_split=0.2, shuffle=True, class_weight={0: 100, 1: 1})"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "history = model.fit(\n",
    "    XX, YY,\n",
    "    epochs=30,\n",
    "    validation_split=0.2, shuffle=True, class_weight={0: 1, 1: 1})"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "model.save_weights(\"temp.h5\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "model.load_weights(\"temp.h5\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "acc = history.history['accuracy']\n",
    "val_acc = history.history['val_accuracy']\n",
    "\n",
    "loss=history.history['loss']\n",
    "val_loss=history.history['val_loss']\n",
    "\n",
    "epochs_range = range(epochs)\n",
    "\n",
    "plt.figure(figsize=(8, 8))\n",
    "plt.subplot(1, 2, 1)\n",
    "plt.plot(epochs_range, acc, label='Training Accuracy')\n",
    "plt.plot(epochs_range, val_acc, label='Validation Accuracy')\n",
    "plt.legend(loc='lower right')\n",
    "plt.title('Training and Validation Accuracy')\n",
    "\n",
    "plt.subplot(1, 2, 2)\n",
    "plt.plot(epochs_range, loss, label='Training Loss')\n",
    "plt.plot(epochs_range, val_loss, label='Validation Loss')\n",
    "plt.legend(loc='upper right')\n",
    "plt.title('Training and Validation Loss')\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "image = val_data_gen.next()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "image = train_data_gen.next()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "print(image[0].shape)\n",
    "predictions = model.predict_classes(image[0])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "print(confusion_matrix(np.argmax(image[1], axis=1), predictions.flatten()))\n",
    "print(classification_report(np.argmax(image[1], axis=1), predictions.flatten()))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "model.save(path.join(PATH, \"model.h5\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "ann = pd.read_csv(path.join(PATH, \"samples\", \"via_export_csv.csv\"))\n",
    "print(ann.head())\n",
    "print(ann[ann.region_count > 0].filename.unique().size)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "grp = ann.groupby(\"filename\")\n",
    "print(grp.mean())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 244,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/suba/.virtualenvs/emotion/lib/python3.6/site-packages/ipykernel_launcher.py:32: TqdmDeprecationWarning: This function will be removed in tqdm==5.0.0\n",
      "Please use `tqdm.notebook.tqdm` instead of `tqdm.tqdm_notebook`\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "1477182bb29545f98d12465083556c9d",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(FloatProgress(value=0.0, max=140.0), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "140/140 [==============================] - 0s 715us/sample\n",
      "12\n"
     ]
    }
   ],
   "source": [
    "# Runs the model through the sample images and draws a bounding box\n",
    "do_nms = True\n",
    "do_prediction = True\n",
    "prob_thresh = 0.9\n",
    "stride = 25\n",
    "width = 50\n",
    "prediction_width = 40\n",
    "do_resize = True\n",
    "base_path = path.join(PATH, \"split\")\n",
    "ann = pd.read_csv(path.join(base_path, \"via_export_csv.csv\"))\n",
    "for file in ann[\"filename\"].unique():\n",
    "    cnt = 0\n",
    "#     if file not in [\"2019_1206_122956_179.JPG\", \"2019_1206_122957_180.JPG\"]:\n",
    "#         continue\n",
    "    if not file.startswith(\"2019_1206_122956_179_3_4\"):\n",
    "        continue\n",
    "    sample_ann = ann[ann[\"filename\"] == file]\n",
    "    training_data = []\n",
    "    img = cv2.imread(path.join(base_path, file))\n",
    "    for _x in range(0, img.shape[1] - width, stride):\n",
    "        for _y in range(0, img.shape[0] - width, stride):\n",
    "            xstart = _x\n",
    "            xend = _x + width\n",
    "            ystart = _y\n",
    "            yend = _y + width\n",
    "            training_data.append([ystart, yend, xstart, xend])\n",
    "    \n",
    "    img = cv2.imread(path.join(base_path, file))\n",
    "    if do_prediction:\n",
    "        all_images = []\n",
    "        predictions = []\n",
    "        for ix in tqdm.tqdm_notebook(range(len(training_data))):\n",
    "            val = training_data[ix]\n",
    "            samp_img = img[val[0]: val[1], val[2]: val[3]]\n",
    "            if do_resize:\n",
    "                samp_img = cv2.resize(samp_img, (prediction_width, prediction_width))\n",
    "            all_images.append(samp_img)\n",
    "            if len(all_images) == 50000:\n",
    "                all_images = np.array(all_images) * 1.0 / 255\n",
    "                temp_predictions = model.predict(all_images, verbose=1).tolist()\n",
    "                predictions.extend(temp_predictions)\n",
    "                all_images = []\n",
    "        all_images = np.array(all_images) * 1.0 / 255\n",
    "        predictions.extend(model.predict(all_images, verbose =1).tolist())\n",
    "\n",
    "    img = cv2.imread(path.join(base_path, \"labels\", file))\n",
    "    all_boxes = []\n",
    "    all_scores = []\n",
    "    for ix in range(len(predictions)):\n",
    "        prob = predictions[ix][1]\n",
    "        if prob > 0.5:\n",
    "            val = training_data[ix]\n",
    "            all_boxes.append([val[2], val[0], width, width])\n",
    "            all_scores.append(prob)\n",
    "    if do_nms:\n",
    "        indices = nms.boxes(all_boxes, all_scores, nms_threshold=0.1)\n",
    "    else:\n",
    "        indices = range(len(all_boxes))\n",
    "     \n",
    "    for ix in indices:\n",
    "        if all_scores[ix] > prob_thresh:\n",
    "            cnt += 1\n",
    "            img = cv2.rectangle(img, (all_boxes[ix][0], all_boxes[ix][1]), (all_boxes[ix][0] + width, all_boxes[ix][1] + width), (0, 255, 0), 2)\n",
    "    cv2.imwrite(path.join(base_path, \"predictions\", file), img)\n",
    "    print(cnt)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 258,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/suba/.virtualenvs/emotion/lib/python3.6/site-packages/ipykernel_launcher.py:38: TqdmDeprecationWarning: This function will be removed in tqdm==5.0.0\n",
      "Please use `tqdm.notebook.tqdm` instead of `tqdm.tqdm_notebook`\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "4b7c7f86f4c345f2bbd247d4742ba433",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(FloatProgress(value=0.0, max=364179.0), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "50000/50000 [==============================] - 30s 605us/sample\n",
      "50000/50000 [==============================] - 29s 590us/sample\n",
      "50000/50000 [==============================] - 29s 581us/sample\n",
      "50000/50000 [==============================] - 29s 577us/sample\n",
      "50000/50000 [==============================] - 30s 591us/sample\n",
      "50000/50000 [==============================] - 30s 595us/sample\n",
      "50000/50000 [==============================] - 30s 591us/sample\n",
      "\n",
      "14179/14179 [==============================] - 8s 595us/sample\n",
      "417\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "3a4182ba617d45bc9290e53fc46c7834",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(FloatProgress(value=0.0, max=364179.0), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "50000/50000 [==============================] - 29s 590us/sample\n",
      "50000/50000 [==============================] - 30s 594us/sample\n",
      "50000/50000 [==============================] - 33s 657us/sample\n",
      "50000/50000 [==============================] - 30s 604us/sample\n",
      "50000/50000 [==============================] - 32s 631us/sample\n",
      "50000/50000 [==============================] - 32s 641us/sample\n",
      "50000/50000 [==============================] - 32s 649us/sample\n",
      "\n",
      "14179/14179 [==============================] - 9s 640us/sample\n",
      "497\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "ae56115e959945ad9d1b3e08af6960c1",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(FloatProgress(value=0.0, max=364179.0), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "50000/50000 [==============================] - 32s 634us/sample\n",
      "50000/50000 [==============================] - 30s 608us/sample\n",
      "50000/50000 [==============================] - 29s 589us/sample\n",
      "50000/50000 [==============================] - 31s 613us/sample\n",
      "50000/50000 [==============================] - 33s 660us/sample\n",
      "50000/50000 [==============================] - 32s 644us/sample\n",
      "50000/50000 [==============================] - 32s 645us/sample\n",
      "\n",
      "14179/14179 [==============================] - 9s 632us/sample\n",
      "453\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "f617edf0bdc341a1b1de0f5b6415db7d",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(FloatProgress(value=0.0, max=364179.0), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "50000/50000 [==============================] - 29s 585us/sample\n",
      "50000/50000 [==============================] - 29s 586us/sample\n",
      "50000/50000 [==============================] - 31s 620us/sample\n",
      "50000/50000 [==============================] - 35s 703us/sample\n",
      "50000/50000 [==============================] - 36s 719us/sample\n",
      "50000/50000 [==============================] - 38s 758us/sample\n",
      "50000/50000 [==============================] - 35s 692us/sample\n",
      "\n",
      "14179/14179 [==============================] - 9s 639us/sample\n",
      "441\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "a4ebe16ca4e04df491b3f880e0ef99aa",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(FloatProgress(value=0.0, max=364179.0), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "50000/50000 [==============================] - 32s 635us/sample\n",
      "50000/50000 [==============================] - 35s 709us/sample\n",
      "50000/50000 [==============================] - 35s 708us/sample\n",
      "50000/50000 [==============================] - 33s 662us/sample\n",
      "50000/50000 [==============================] - 32s 648us/sample\n",
      "50000/50000 [==============================] - 31s 629us/sample\n",
      "50000/50000 [==============================] - 31s 613us/sample\n",
      "\n",
      "14179/14179 [==============================] - 8s 594us/sample\n",
      "363\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "dca9e37261d144b9986b9541988599bf",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(FloatProgress(value=0.0, max=364179.0), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "50000/50000 [==============================] - 32s 648us/sample\n",
      "50000/50000 [==============================] - 33s 657us/sample\n",
      "50000/50000 [==============================] - 33s 666us/sample\n",
      "50000/50000 [==============================] - 32s 648us/sample\n",
      "50000/50000 [==============================] - 30s 607us/sample\n",
      "50000/50000 [==============================] - 30s 597us/sample\n",
      "50000/50000 [==============================] - 31s 621us/sample\n",
      "\n",
      "14179/14179 [==============================] - 9s 668us/sample\n",
      "352\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "1e406dc1d1034273bfbd5ab5fce34f68",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(FloatProgress(value=0.0, max=364179.0), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "50000/50000 [==============================] - 33s 662us/sample\n",
      "50000/50000 [==============================] - 32s 634us/sample\n",
      "50000/50000 [==============================] - 31s 623us/sample\n",
      "50000/50000 [==============================] - 29s 584us/sample\n",
      "50000/50000 [==============================] - 29s 577us/sample\n",
      "50000/50000 [==============================] - 30s 601us/sample\n",
      "50000/50000 [==============================] - 31s 622us/sample\n",
      "\n",
      "14179/14179 [==============================] - 10s 692us/sample\n",
      "378\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "d5afeb85f3fb46e7b748c66aabddd93f",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(FloatProgress(value=0.0, max=364179.0), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "50000/50000 [==============================] - 32s 632us/sample\n",
      "50000/50000 [==============================] - 33s 660us/sample\n",
      "50000/50000 [==============================] - 29s 589us/sample\n",
      "50000/50000 [==============================] - 29s 576us/sample\n",
      "50000/50000 [==============================] - 30s 591us/sample\n",
      "50000/50000 [==============================] - 31s 622us/sample\n",
      "50000/50000 [==============================] - 32s 636us/sample\n",
      "\n",
      "14179/14179 [==============================] - 9s 622us/sample\n",
      "334\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "42b7c95b71ac43748c49a18840d647dd",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(FloatProgress(value=0.0, max=364179.0), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "50000/50000 [==============================] - 31s 624us/sample\n",
      "50000/50000 [==============================] - 30s 594us/sample\n",
      "50000/50000 [==============================] - 29s 575us/sample\n",
      "50000/50000 [==============================] - 29s 586us/sample\n",
      "50000/50000 [==============================] - 31s 621us/sample\n",
      "50000/50000 [==============================] - 32s 635us/sample\n",
      "50000/50000 [==============================] - 32s 634us/sample\n",
      "\n",
      "14179/14179 [==============================] - 9s 624us/sample\n",
      "337\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "38f8802f536e4db5b38078567dddd851",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(FloatProgress(value=0.0, max=364179.0), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "50000/50000 [==============================] - 30s 592us/sample\n",
      "50000/50000 [==============================] - 32s 633us/sample\n",
      "50000/50000 [==============================] - 35s 699us/sample\n",
      "50000/50000 [==============================] - 31s 628us/sample\n",
      "50000/50000 [==============================] - 36s 711us/sample\n",
      "50000/50000 [==============================] - 38s 758us/sample\n",
      "50000/50000 [==============================] - 39s 787us/sample\n",
      "\n",
      "14179/14179 [==============================] - 9s 619us/sample\n",
      "356\n"
     ]
    }
   ],
   "source": [
    "# Runs the model through the sample images and draws a bounding box\n",
    "do_nms = True\n",
    "do_prediction = True\n",
    "prob_thresh = 0.99\n",
    "stride = 30\n",
    "width = 60\n",
    "prediction_width = 40\n",
    "do_resize = True\n",
    "base_path = path.join(PATH, \"Set2\")\n",
    "ann = pd.read_csv(path.join(base_path, \"via_export_csv.csv\"))\n",
    "for file in ann[\"filename\"].unique():\n",
    "    cnt = 0\n",
    "#     if file not in [\"2019_1206_122956_179_3_4.jpg\", \"2019_1206_122956_179_3_3.jpg\", \"2019_1206_122956_179_8_3.jpg\",\n",
    "#                    \"2019_1206_122956_179_6_2.jpg\"]:\n",
    "#         continue\n",
    "#     if not file.startswith(\"2019_1206_122954_178\"):\n",
    "#         continue\n",
    "    sample_ann = ann[ann[\"filename\"] == file]\n",
    "    training_data = []\n",
    "    img = cv2.imread(path.join(base_path, file))\n",
    "    widths = [width - 20, width - 10, width, width + 10, width + 20]\n",
    "    for w in widths:\n",
    "        for h in widths:\n",
    "            stridew = int(w/2)\n",
    "            strideh = int(h/2)\n",
    "            for _x in range(0, img.shape[1] - w, stridew):\n",
    "                for _y in range(0, img.shape[0] - h, strideh):\n",
    "                    xstart = _x\n",
    "                    xend = _x + width\n",
    "                    ystart = _y\n",
    "                    yend = _y + width\n",
    "                    training_data.append([ystart, yend, xstart, xend])\n",
    "    \n",
    "    img = cv2.imread(path.join(base_path, file))\n",
    "    if do_prediction:\n",
    "        all_images = []\n",
    "        predictions = []\n",
    "        for ix in tqdm.tqdm_notebook(range(len(training_data))):\n",
    "            val = training_data[ix]\n",
    "            samp_img = img[val[0]: val[1], val[2]: val[3]]\n",
    "            if do_resize:\n",
    "                samp_img = cv2.resize(samp_img, (prediction_width, prediction_width))\n",
    "            all_images.append(samp_img)\n",
    "            if len(all_images) == 50000:\n",
    "                all_images = np.array(all_images) * 1.0 / 255\n",
    "                temp_predictions = model.predict(all_images, verbose=1).tolist()\n",
    "                predictions.extend(temp_predictions)\n",
    "                all_images = []\n",
    "        all_images = np.array(all_images) * 1.0 / 255\n",
    "        predictions.extend(model.predict(all_images, verbose =1).tolist())\n",
    "\n",
    "    img = cv2.imread(path.join(base_path, \"labels\", file))\n",
    "    all_boxes = []\n",
    "    all_scores = []\n",
    "    for ix in range(len(predictions)):\n",
    "        prob = predictions[ix][1]\n",
    "        if prob > 0.5:\n",
    "            val = training_data[ix]\n",
    "            all_boxes.append([val[2], val[0], width, width])\n",
    "            all_scores.append(prob)\n",
    "    if do_nms:\n",
    "        indices = nms.boxes(all_boxes, all_scores, nms_threshold=0.1)\n",
    "    else:\n",
    "        indices = range(len(all_boxes))\n",
    "     \n",
    "    for ix in indices:\n",
    "        if all_scores[ix] > prob_thresh:\n",
    "            cnt += 1\n",
    "            img = cv2.rectangle(img, (all_boxes[ix][0], all_boxes[ix][1]), (all_boxes[ix][0] + width, all_boxes[ix][1] + width), (0, 255, 0), 2)\n",
    "    cv2.imwrite(path.join(base_path, \"predictions\", file), img)\n",
    "    print(cnt)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 218,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[[0.999998927116394, 1.0273620318912435e-06],\n",
       " [0.9999948740005493, 5.125348707224475e-06],\n",
       " [0.9999912977218628, 8.733434697205666e-06],\n",
       " [0.999976396560669, 2.364775355090387e-05],\n",
       " [0.9999991655349731, 8.51351103392517e-07],\n",
       " [0.9999991655349731, 8.226190857385518e-07],\n",
       " [0.9999992847442627, 7.218312134682492e-07],\n",
       " [0.9999988079071045, 1.237756237060239e-06],\n",
       " [0.9999985694885254, 1.3760630963588483e-06],\n",
       " [0.9999985694885254, 1.4730513839822379e-06],\n",
       " [0.9999986886978149, 1.3324040537554538e-06],\n",
       " [0.9999986886978149, 1.3554413271776866e-06],\n",
       " [0.9999990463256836, 9.257672104467929e-07],\n",
       " [0.9999992847442627, 7.035993121462525e-07],\n",
       " [0.999998927116394, 1.0843243671843084e-06],\n",
       " [0.9999985694885254, 1.4032702893018723e-06],\n",
       " [0.9999988079071045, 1.2148790347055183e-06],\n",
       " [0.9999992847442627, 6.980885700613726e-07],\n",
       " [0.9999990463256836, 9.086668910640583e-07],\n",
       " [0.9999991655349731, 8.277697816083673e-07],\n",
       " [0.9999992847442627, 6.864684678475896e-07],\n",
       " [0.9999939203262329, 6.104091880843043e-06],\n",
       " [0.9999938011169434, 6.1828395701013505e-06],\n",
       " [0.9999942779541016, 5.758285624324344e-06],\n",
       " [0.9999992847442627, 7.709242026976426e-07],\n",
       " [0.9999911785125732, 8.840511327434797e-06],\n",
       " [0.9999992847442627, 7.135740247576905e-07],\n",
       " [0.9999921321868896, 7.85606971476227e-06],\n",
       " [0.9999985694885254, 1.3739321502725943e-06],\n",
       " [0.9999992847442627, 7.359880669355334e-07],\n",
       " [0.9999992847442627, 7.324827606680628e-07],\n",
       " [0.9999990463256836, 8.975865171123587e-07],\n",
       " [0.9999990463256836, 9.398474958288716e-07],\n",
       " [0.9999991655349731, 8.47691580929677e-07],\n",
       " [0.9999991655349731, 8.852777000356582e-07],\n",
       " [0.9999939203262329, 6.0909214880666696e-06],\n",
       " [0.9999938011169434, 6.169328116811812e-06],\n",
       " [0.9999364614486694, 6.35117685305886e-05],\n",
       " [0.9998548030853271, 0.00014516685041598976],\n",
       " [0.9999788999557495, 2.107361979142297e-05],\n",
       " [0.9999938011169434, 6.253239462239435e-06],\n",
       " [0.9999938011169434, 6.21033041170449e-06],\n",
       " [0.9998972415924072, 0.0001027681864798069],\n",
       " [0.9999504089355469, 4.958073623129167e-05],\n",
       " [0.9999936819076538, 6.317302904790267e-06],\n",
       " [0.9999942779541016, 5.7029346862691455e-06],\n",
       " [0.9999977350234985, 2.220761416538153e-06],\n",
       " [0.9999949932098389, 4.961068498232635e-06],\n",
       " [0.9999992847442627, 6.919838710928161e-07],\n",
       " [0.9999992847442627, 7.392421821350581e-07],\n",
       " [0.9999992847442627, 7.626624096701562e-07],\n",
       " [0.9999897480010986, 1.0294303137925453e-05],\n",
       " [0.999983549118042, 1.6406656868639402e-05],\n",
       " [0.9999902248382568, 9.74390877672704e-06],\n",
       " [0.9999939203262329, 6.138329808891285e-06],\n",
       " [0.9999982118606567, 1.7545792161399731e-06],\n",
       " [0.9999959468841553, 4.112117494514678e-06],\n",
       " [0.9999927282333374, 7.27596489014104e-06],\n",
       " [0.9999992847442627, 7.136080739655881e-07],\n",
       " [0.9999992847442627, 7.511858370889968e-07],\n",
       " [0.999992847442627, 7.109166745067341e-06],\n",
       " [0.9999992847442627, 6.874793712086102e-07],\n",
       " [0.999998927116394, 1.0253359050693689e-06],\n",
       " [0.9999991655349731, 8.588567652623169e-07],\n",
       " [1.0, 0.0],\n",
       " [0.9999988079071045, 1.198821223624691e-06],\n",
       " [0.9999938011169434, 6.1963219195604324e-06],\n",
       " [0.9999963045120239, 3.6472620195127092e-06],\n",
       " [0.9999985694885254, 1.486469159317494e-06],\n",
       " [0.9999986886978149, 1.2860884908150183e-06],\n",
       " [0.9999990463256836, 9.151729045697721e-07],\n",
       " [0.9999986886978149, 1.2910605846627732e-06],\n",
       " [0.9999986886978149, 1.2778142490788014e-06],\n",
       " [0.9999985694885254, 1.3787007446808275e-06],\n",
       " [0.999998927116394, 1.112228801503079e-06],\n",
       " [0.9999985694885254, 1.381852030135633e-06],\n",
       " [0.9999985694885254, 1.405151806466165e-06],\n",
       " [0.9999986886978149, 1.3044865454503451e-06],\n",
       " [0.9999977350234985, 2.246600388389197e-06],\n",
       " [0.9999939203262329, 6.118003511801362e-06],\n",
       " [0.9999732971191406, 2.6684354452299885e-05],\n",
       " [0.999998927116394, 1.0626481525832787e-06],\n",
       " [0.9999986886978149, 1.3053253269390552e-06],\n",
       " [0.9999992847442627, 7.035516773612471e-07],\n",
       " [0.9999992847442627, 7.520158646912023e-07],\n",
       " [0.9999852180480957, 1.4741511222382542e-05],\n",
       " [0.9999755620956421, 2.4402233975706622e-05],\n",
       " [0.9999910593032837, 8.978803634818178e-06],\n",
       " [0.9999986886978149, 1.300462031395e-06],\n",
       " [0.999993085861206, 6.876603947603144e-06],\n",
       " [0.9999969005584717, 3.1058339118317235e-06],\n",
       " [0.9999970197677612, 2.9219941097835544e-06],\n",
       " [0.9999971389770508, 2.802170683935401e-06],\n",
       " [0.9999990463256836, 9.803027296584332e-07],\n",
       " [0.9999992847442627, 7.054894126667932e-07],\n",
       " [0.9999985694885254, 1.4440711311181076e-06],\n",
       " [0.9999988079071045, 1.136778223553847e-06],\n",
       " [0.9999988079071045, 1.146123622675077e-06],\n",
       " [0.9999990463256836, 1.0105676437888178e-06],\n",
       " [0.999994158744812, 5.864648755959934e-06],\n",
       " [0.9999992847442627, 7.612054560013348e-07],\n",
       " [0.999996542930603, 3.4740480714390287e-06],\n",
       " [0.9999938011169434, 6.1559162531921174e-06],\n",
       " [0.9999992847442627, 7.146555276449362e-07],\n",
       " [0.9999990463256836, 9.210691018779471e-07],\n",
       " [0.9999984502792358, 1.4977132423155126e-06],\n",
       " [0.999998927116394, 1.037754714161565e-06],\n",
       " [0.9999938011169434, 6.216896053956589e-06],\n",
       " [0.9999991655349731, 8.5924182258168e-07],\n",
       " [0.999998927116394, 1.039020276039082e-06],\n",
       " [0.9999990463256836, 9.194314429805672e-07],\n",
       " [0.9999939203262329, 6.112567916716216e-06],\n",
       " [0.999972939491272, 2.708564898057375e-05],\n",
       " [0.9999938011169434, 6.216184829099802e-06],\n",
       " [0.9999992847442627, 7.329915092668671e-07],\n",
       " [0.9999990463256836, 9.754645589055144e-07],\n",
       " [0.9999991655349731, 7.996436579560395e-07],\n",
       " [0.9999992847442627, 7.054846378196089e-07],\n",
       " [0.9999939203262329, 6.102922270656563e-06],\n",
       " [0.9999841451644897, 1.5861507563386112e-05],\n",
       " [0.9999991655349731, 8.899341992218979e-07],\n",
       " [0.9999922513961792, 7.786709829815663e-06],\n",
       " [0.9999936819076538, 6.265069714572746e-06],\n",
       " [0.9999855756759644, 1.4422019376070239e-05],\n",
       " [0.9999991655349731, 8.050089945754735e-07],\n",
       " [0.9999990463256836, 9.401423994859215e-07],\n",
       " [0.9999986886978149, 1.3179464986023959e-06],\n",
       " [0.9999985694885254, 1.4220136108633596e-06],\n",
       " [0.9999984502792358, 1.552532125970174e-06],\n",
       " [0.9999985694885254, 1.479479692534369e-06],\n",
       " [0.9999986886978149, 1.3671788110514171e-06],\n",
       " [0.9999984502792358, 1.5434529814228881e-06],\n",
       " [0.9999992847442627, 6.896976856296533e-07],\n",
       " [0.9999992847442627, 7.611198498125304e-07],\n",
       " [0.9999991655349731, 8.810487770460895e-07],\n",
       " [0.9999991655349731, 8.472349577459681e-07],\n",
       " [0.9999992847442627, 7.191611075541005e-07],\n",
       " [0.9999990463256836, 9.371937039759359e-07],\n",
       " [0.9999990463256836, 9.131433102993469e-07],\n",
       " [0.9999990463256836, 9.569322401148384e-07],\n",
       " [0.9999992847442627, 6.987499432398181e-07],\n",
       " [0.9999924898147583, 7.471009666915052e-06],\n",
       " [0.9999963045120239, 3.676435653687804e-06],\n",
       " [0.9999991655349731, 8.56867927723215e-07],\n",
       " [0.9999988079071045, 1.1843440006487072e-06],\n",
       " [0.9999990463256836, 9.795402320378344e-07],\n",
       " [0.9999991655349731, 8.123214456645655e-07],\n",
       " [0.9999431371688843, 5.686001168214716e-05],\n",
       " [0.9999991655349731, 8.756208558224898e-07],\n",
       " [0.9999990463256836, 9.25962353903742e-07],\n",
       " [0.9999984502792358, 1.5596909861415043e-06],\n",
       " [0.9999982118606567, 1.8282306655237335e-06],\n",
       " [0.9999992847442627, 7.167249691519828e-07],\n",
       " [0.9999971389770508, 2.819043857016368e-06],\n",
       " [0.9999992847442627, 7.712161504969117e-07],\n",
       " [0.999998927116394, 1.1034038607249386e-06],\n",
       " [0.9999990463256836, 9.508336233921e-07],\n",
       " [0.9999990463256836, 9.879988738248358e-07],\n",
       " [0.9999990463256836, 9.425285725228605e-07],\n",
       " [0.9999991655349731, 8.070913963820203e-07],\n",
       " [0.9999938011169434, 6.176533588586608e-06],\n",
       " [0.9999949932098389, 4.986312433175044e-06],\n",
       " [0.999998927116394, 1.05367792002653e-06],\n",
       " [0.9999992847442627, 6.852440606053278e-07],\n",
       " [0.999998927116394, 1.013982000586111e-06],\n",
       " [0.9999943971633911, 5.5490663726232015e-06],\n",
       " [0.9999992847442627, 7.591552844132821e-07],\n",
       " [0.9999938011169434, 6.231682618818013e-06],\n",
       " [0.9999957084655762, 4.336171059549088e-06],\n",
       " [0.9999654293060303, 3.4575281461002305e-05],\n",
       " [0.999971866607666, 2.8085221856599674e-05],\n",
       " [0.9999939203262329, 6.117706107033882e-06],\n",
       " [0.9999988079071045, 1.179499463432876e-06],\n",
       " [0.9999938011169434, 6.166504590510158e-06],\n",
       " [0.9999951124191284, 4.90820229970268e-06],\n",
       " [0.9999939203262329, 6.118423698353581e-06],\n",
       " [0.9999986886978149, 1.312146196141839e-06],\n",
       " [0.9999990463256836, 9.316224236499693e-07],\n",
       " [0.9999992847442627, 7.544581990259758e-07],\n",
       " [0.999998927116394, 1.0283765732310712e-06],\n",
       " [0.9999988079071045, 1.2127547961426899e-06],\n",
       " [0.9999306201934814, 6.937197031220421e-05],\n",
       " [0.9999618530273438, 3.820455094682984e-05],\n",
       " [0.9999839067459106, 1.6148911527125165e-05],\n",
       " [0.9999992847442627, 7.367999046437035e-07],\n",
       " [0.9999991655349731, 8.20626496533805e-07],\n",
       " [0.9999922513961792, 7.695448402955662e-06],\n",
       " [0.9999852180480957, 1.4762262253498193e-05],\n",
       " [0.9999641180038452, 3.592623397707939e-05],\n",
       " [0.9999594688415527, 4.056334728375077e-05],\n",
       " [0.9999018907546997, 9.805487934499979e-05],\n",
       " [0.9999219179153442, 7.805793575244024e-05],\n",
       " [0.9999948740005493, 5.114343366585672e-06],\n",
       " [0.9999947547912598, 5.212658379605273e-06],\n",
       " [0.9999933242797852, 6.63935452394071e-06],\n",
       " [0.9999784231185913, 2.162746932299342e-05],\n",
       " [0.9999773502349854, 2.266790033900179e-05],\n",
       " [0.999967098236084, 3.295440910733305e-05],\n",
       " [0.9999750852584839, 2.4860219127731398e-05],\n",
       " [0.9999946355819702, 5.385053555073682e-06],\n",
       " [0.9999892711639404, 1.0679619663278572e-05],\n",
       " [0.9999440908432007, 5.587921987171285e-05],\n",
       " [0.9999914169311523, 8.64103640196845e-06],\n",
       " [0.9999991655349731, 7.777564405841986e-07],\n",
       " [0.9999872446060181, 1.277658884646371e-05],\n",
       " [0.9999986886978149, 1.269653694180306e-06],\n",
       " [0.999998927116394, 1.0161930958929588e-06],\n",
       " [0.9999912977218628, 8.736008567211684e-06],\n",
       " [0.9999885559082031, 1.1499333595565986e-05],\n",
       " [0.9999992847442627, 7.540259048255393e-07],\n",
       " [0.9999990463256836, 9.463167884860013e-07],\n",
       " [0.9999991655349731, 8.819314416541602e-07],\n",
       " [1.0, 0.0],\n",
       " [1.0, 0.0],\n",
       " [0.9999983310699463, 1.6306464658555342e-06],\n",
       " [0.9999992847442627, 6.975880637583032e-07],\n",
       " [0.9999979734420776, 1.9871915810654173e-06],\n",
       " [0.9999986886978149, 1.3059328693998395e-06],\n",
       " [0.9999958276748657, 4.153686859353911e-06],\n",
       " [0.9999936819076538, 6.312942787189968e-06],\n",
       " [0.9999986886978149, 1.2698800446742098e-06],\n",
       " [0.9999985694885254, 1.3926421615906293e-06],\n",
       " [0.9999990463256836, 9.774815907803713e-07],\n",
       " [0.999998927116394, 1.0752419257187285e-06],\n",
       " [0.9999985694885254, 1.4549418665410485e-06],\n",
       " [0.9999988079071045, 1.174190856545465e-06],\n",
       " [0.999998927116394, 1.0714647942222655e-06],\n",
       " [0.9999938011169434, 6.252011189644691e-06],\n",
       " [0.9999992847442627, 7.471533649550111e-07],\n",
       " [0.9999990463256836, 9.776418892215588e-07],\n",
       " [0.9999991655349731, 8.097666750472854e-07],\n",
       " [0.9999979734420776, 1.98224006453529e-06],\n",
       " [0.9999992847442627, 7.168781053223938e-07],\n",
       " [0.999971866607666, 2.8154490792076103e-05],\n",
       " [0.9998051524162292, 0.0001949220459209755],\n",
       " [0.9997761845588684, 0.00022387078206520528],\n",
       " [0.9999992847442627, 6.846431119811314e-07],\n",
       " [0.9999336004257202, 6.634012243011966e-05],\n",
       " [0.9999537467956543, 4.6244196710176766e-05],\n",
       " [0.9999773502349854, 2.2643915144726634e-05],\n",
       " [0.9999195337295532, 8.046218135859817e-05],\n",
       " [0.9999992847442627, 7.339750140999968e-07],\n",
       " [0.999998927116394, 1.1310880836390425e-06],\n",
       " [0.9999991655349731, 8.910977271625597e-07],\n",
       " [0.9999992847442627, 6.939426384633407e-07],\n",
       " [0.9999970197677612, 2.937492354249116e-06],\n",
       " [0.9999960660934448, 3.990884124505101e-06],\n",
       " [0.9999992847442627, 7.008194984337024e-07],\n",
       " [0.9999985694885254, 1.451091065973742e-06],\n",
       " [0.9999992847442627, 7.693179782108928e-07],\n",
       " [0.9999988079071045, 1.1854672266053967e-06],\n",
       " [0.9999990463256836, 9.277903814108868e-07],\n",
       " [0.9999990463256836, 9.343299325337284e-07],\n",
       " [0.9999992847442627, 7.574558367196005e-07],\n",
       " [0.9999990463256836, 9.785970860320958e-07],\n",
       " [0.9999939203262329, 6.091293016652344e-06],\n",
       " [0.9999990463256836, 9.860184491117252e-07],\n",
       " [0.999998927116394, 1.0758071766758803e-06],\n",
       " [0.9999992847442627, 7.447368375324004e-07],\n",
       " [0.9999938011169434, 6.227114226931008e-06],\n",
       " [0.9999936819076538, 6.288121312536532e-06],\n",
       " [0.9999939203262329, 6.127024789748248e-06],\n",
       " [0.9999988079071045, 1.1381190461179358e-06],\n",
       " [0.9999988079071045, 1.176678324554814e-06],\n",
       " [0.999997615814209, 2.39570499616093e-06],\n",
       " [0.9999901056289673, 9.890161891235039e-06],\n",
       " [0.9999914169311523, 8.5270758063416e-06],\n",
       " [0.9999948740005493, 5.079728907730896e-06],\n",
       " [0.999998927116394, 1.0986200322804507e-06],\n",
       " [0.9999991655349731, 7.865004363338812e-07],\n",
       " [0.9999977350234985, 2.211255377915222e-06],\n",
       " [0.9999934434890747, 6.514056167361559e-06],\n",
       " [0.9999991655349731, 7.910612680461782e-07],\n",
       " [0.9999990463256836, 9.407666539118509e-07],\n",
       " [0.9999992847442627, 7.456343951162125e-07],\n",
       " [0.9999988079071045, 1.1921243867618614e-06],\n",
       " [0.9999988079071045, 1.1802939070548746e-06],\n",
       " [0.9999990463256836, 9.942685892383452e-07],\n",
       " [0.9999986886978149, 1.347978809462802e-06],\n",
       " [0.9999990463256836, 9.371614737574419e-07],\n",
       " [0.9999911785125732, 8.81697724253172e-06],\n",
       " [0.9999924898147583, 7.541268587374361e-06],\n",
       " [0.9999991655349731, 8.610982149548363e-07],\n",
       " [0.9999896287918091, 1.0315274266758934e-05],\n",
       " [0.9999939203262329, 6.117420070950175e-06],\n",
       " [0.9999990463256836, 1.008985464068246e-06],\n",
       " [0.9999985694885254, 1.37849303882831e-06],\n",
       " [0.9999992847442627, 7.74287059357448e-07],\n",
       " [0.9999557733535767, 4.424632788868621e-05],\n",
       " [0.9998020529747009, 0.0001979349326575175],\n",
       " [0.9999967813491821, 3.2398870644101407e-06],\n",
       " [0.9999991655349731, 8.811126122054702e-07],\n",
       " [0.9999986886978149, 1.3569662087320467e-06],\n",
       " [0.9999991655349731, 8.737706025385705e-07],\n",
       " [0.9999805688858032, 1.9403150872676633e-05],\n",
       " [0.9999456405639648, 5.433883052319288e-05],\n",
       " [0.9999992847442627, 7.007446356510627e-07],\n",
       " [0.9999992847442627, 7.323835689021507e-07],\n",
       " [0.999998927116394, 1.1158972483826801e-06],\n",
       " [0.9999990463256836, 9.051320830621989e-07],\n",
       " [0.9999990463256836, 9.593984486855334e-07],\n",
       " [0.9999990463256836, 9.779515721675125e-07],\n",
       " [0.9999984502792358, 1.5045418422232615e-06],\n",
       " [0.999998927116394, 1.1145444887006306e-06],\n",
       " [0.999998927116394, 1.0550463684921851e-06],\n",
       " [0.9999977350234985, 2.3114737359719584e-06],\n",
       " [0.9999938011169434, 6.236076387722278e-06],\n",
       " [0.9999983310699463, 1.6959436379693216e-06],\n",
       " [0.9999991655349731, 8.267946327578102e-07],\n",
       " [0.999998927116394, 1.0709539992603823e-06],\n",
       " [0.9999985694885254, 1.3998031818118761e-06],\n",
       " [0.9999986886978149, 1.259170176126645e-06],\n",
       " [0.9999992847442627, 7.337762326642405e-07],\n",
       " [0.9999901056289673, 9.87201292446116e-06],\n",
       " [0.9999992847442627, 7.290880148502765e-07],\n",
       " [0.999998927116394, 1.0909316188190132e-06],\n",
       " [0.9999990463256836, 9.977326271837228e-07],\n",
       " [0.9999467134475708, 5.332641376298852e-05],\n",
       " [0.9998413324356079, 0.00015869733761064708],\n",
       " [0.9999876022338867, 1.2351606528682169e-05],\n",
       " [0.9999980926513672, 1.914644826683798e-06],\n",
       " [0.9999990463256836, 1.004103296509129e-06],\n",
       " [0.9999992847442627, 7.356217679443944e-07],\n",
       " [0.9999939203262329, 6.0822262639703695e-06],\n",
       " [0.9999992847442627, 7.298456239368534e-07],\n",
       " [0.9999991655349731, 8.28640168037964e-07],\n",
       " [0.9999991655349731, 8.315415129800385e-07],\n",
       " [0.999994158744812, 5.820885235152673e-06],\n",
       " [0.9999992847442627, 6.91762181759259e-07],\n",
       " [0.9999938011169434, 6.199868039402645e-06],\n",
       " [0.9999939203262329, 6.103201485530008e-06],\n",
       " [0.9999991655349731, 8.130872402034584e-07],\n",
       " [0.9999992847442627, 7.070506740092242e-07],\n",
       " [0.9999992847442627, 7.632146434843889e-07],\n",
       " [0.9999992847442627, 7.718709866821882e-07],\n",
       " [0.999933123588562, 6.690632289974019e-05],\n",
       " [0.9999871253967285, 1.2829382285417523e-05],\n",
       " [0.9999991655349731, 8.260529966719332e-07],\n",
       " [0.9999991655349731, 8.438659051535069e-07],\n",
       " [0.9999955892562866, 4.4568328121386e-06],\n",
       " [0.9999992847442627, 6.909327225912421e-07],\n",
       " [0.9999978542327881, 2.158458755729953e-06],\n",
       " [0.9999885559082031, 1.1496932529553305e-05],\n",
       " [0.9999932050704956, 6.790653060306795e-06],\n",
       " [0.9999938011169434, 6.244479209271958e-06],\n",
       " [0.9999480247497559, 5.1960589189548045e-05],\n",
       " [0.9993426203727722, 0.0006573124555870891],\n",
       " [0.9999579191207886, 4.2043524445034564e-05],\n",
       " [0.9999958276748657, 4.203073785902234e-06],\n",
       " [0.9999933242797852, 6.658402071479941e-06],\n",
       " [0.9999936819076538, 6.267041953833541e-06],\n",
       " [0.9999938011169434, 6.149942691990873e-06],\n",
       " [0.9999936819076538, 6.278551609284477e-06],\n",
       " [0.9999963045120239, 3.7424740639835363e-06],\n",
       " [0.9999991655349731, 8.202571848414664e-07],\n",
       " [0.9999938011169434, 6.254838353925152e-06],\n",
       " [0.9999992847442627, 7.106572184056859e-07],\n",
       " [0.999998927116394, 1.0701882047214895e-06],\n",
       " [0.9999990463256836, 9.499806878920936e-07],\n",
       " [1.0, 0.0],\n",
       " [1.0, 0.0],\n",
       " [1.0, 0.0],\n",
       " [1.0, 0.0],\n",
       " [0.9999992847442627, 7.267617547768168e-07],\n",
       " [0.9999896287918091, 1.03798420241219e-05],\n",
       " [0.9999990463256836, 9.388226658302301e-07],\n",
       " [0.9999992847442627, 7.38536130029388e-07],\n",
       " [0.9999991655349731, 7.964449082464853e-07],\n",
       " [0.999998927116394, 1.1062402336392552e-06],\n",
       " [0.9999988079071045, 1.2127663922001375e-06],\n",
       " [0.9999991655349731, 8.686302521709877e-07],\n",
       " [0.999998927116394, 1.1321909596517798e-06],\n",
       " [0.9999988079071045, 1.21319089885219e-06],\n",
       " [0.9999990463256836, 9.587691920387442e-07],\n",
       " [0.9999986886978149, 1.3009544090891723e-06],\n",
       " [0.9999991655349731, 8.296958071696281e-07],\n",
       " [0.9999991655349731, 8.017696018214338e-07],\n",
       " [0.9999991655349731, 8.497353292113985e-07],\n",
       " [0.9999936819076538, 6.275653959164629e-06],\n",
       " [0.9999938011169434, 6.143670361780096e-06],\n",
       " [0.999998927116394, 1.1074394024035428e-06],\n",
       " [0.9999939203262329, 6.084175311116269e-06],\n",
       " [0.9994563460350037, 0.000543684174772352],\n",
       " [0.9999251365661621, 7.487828406738117e-05],\n",
       " [0.9999992847442627, 7.266564239216677e-07],\n",
       " [0.9999936819076538, 6.314797246886883e-06],\n",
       " [0.9999825954437256, 1.7369038687320426e-05],\n",
       " [0.9999966621398926, 3.3006449484673794e-06],\n",
       " [0.9999723434448242, 2.7658594262902625e-05],\n",
       " [0.9999988079071045, 1.1697916306729894e-06],\n",
       " [0.9999991655349731, 8.734165817259054e-07],\n",
       " [0.999997615814209, 2.3268005406862358e-06],\n",
       " [0.9999992847442627, 6.919983661646256e-07],\n",
       " [0.9999909400939941, 9.059663170774002e-06],\n",
       " [0.9999991655349731, 8.466566328024783e-07],\n",
       " [0.9999990463256836, 9.558058309266926e-07],\n",
       " [0.9999990463256836, 9.216683451995777e-07],\n",
       " [0.999998927116394, 1.064380398929643e-06],\n",
       " [0.9999990463256836, 9.621170420359704e-07],\n",
       " [0.9999990463256836, 9.460839578423474e-07],\n",
       " [0.9999986886978149, 1.2748199651468894e-06],\n",
       " [0.999998927116394, 1.1304399549771915e-06],\n",
       " [0.999998927116394, 1.0449438150317292e-06],\n",
       " [0.9999991655349731, 8.817195293886471e-07],\n",
       " [0.9999992847442627, 7.382446369774698e-07],\n",
       " [0.9999992847442627, 6.911679975019069e-07],\n",
       " [0.9999991655349731, 8.582655937061645e-07],\n",
       " [0.9999991655349731, 8.040943839660031e-07],\n",
       " [0.9999991655349731, 7.991740744728304e-07],\n",
       " [0.999998927116394, 1.094293111236766e-06],\n",
       " [0.9999991655349731, 8.28145630293875e-07],\n",
       " [0.9999939203262329, 6.115051746746758e-06],\n",
       " [0.9999953508377075, 4.662228093366139e-06],\n",
       " [0.999998927116394, 1.0593576007522643e-06],\n",
       " [0.9999992847442627, 6.857167136331555e-07],\n",
       " [0.9999992847442627, 7.560845460830024e-07],\n",
       " [0.9999992847442627, 6.941398851267877e-07],\n",
       " [0.9999990463256836, 9.146092452283483e-07],\n",
       " [0.9999990463256836, 9.23038783184893e-07],\n",
       " [0.9999991655349731, 8.803323225947679e-07],\n",
       " [0.9999990463256836, 9.103339380089892e-07],\n",
       " [0.9999986886978149, 1.3044865454503451e-06],\n",
       " [0.999998927116394, 1.0319713510398287e-06],\n",
       " [0.9999990463256836, 9.882345466394327e-07],\n",
       " [0.9999986886978149, 1.305225737269211e-06],\n",
       " [0.999998927116394, 1.096986807169742e-06],\n",
       " [0.9999986886978149, 1.2856936564276111e-06],\n",
       " [0.9999990463256836, 9.075055231733131e-07],\n",
       " [0.9999991655349731, 7.76461718032806e-07],\n",
       " [0.9999990463256836, 9.75429202298983e-07],\n",
       " [0.999998927116394, 1.0837020454346202e-06],\n",
       " [0.999998927116394, 1.0307556976840715e-06],\n",
       " [0.999998927116394, 1.1230641803194885e-06],\n",
       " [0.9999991655349731, 8.485384341838653e-07],\n",
       " [0.9999982118606567, 1.8377050992057775e-06],\n",
       " [0.9998652935028076, 0.00013472016144078225],\n",
       " [0.9999867677688599, 1.3277007383294404e-05],\n",
       " [0.9998679161071777, 0.00013210621546022594],\n",
       " [0.9999891519546509, 1.0849681530089583e-05],\n",
       " [0.9999988079071045, 1.1724418982339557e-06],\n",
       " [0.9999990463256836, 9.209742302118684e-07],\n",
       " [0.9999936819076538, 6.264185230975272e-06],\n",
       " [0.9999926090240479, 7.396734417852713e-06],\n",
       " [0.9999991655349731, 8.441782028967282e-07],\n",
       " [0.9999992847442627, 7.149636758185807e-07],\n",
       " [0.9999938011169434, 6.245610620680964e-06],\n",
       " [0.9999961853027344, 3.830529749393463e-06],\n",
       " [0.9999991655349731, 8.33010687983915e-07],\n",
       " [0.9999939203262329, 6.115191808930831e-06],\n",
       " [0.9999992847442627, 7.264596320055716e-07],\n",
       " [0.9999990463256836, 9.913892426993698e-07],\n",
       " [0.9999991655349731, 8.910960218599939e-07],\n",
       " [0.9999992847442627, 6.980672537792998e-07],\n",
       " [0.9999991655349731, 7.992319979166496e-07],\n",
       " [0.9999939203262329, 6.098424819356296e-06],\n",
       " [0.9999971389770508, 2.8240592655492947e-06],\n",
       " [0.9999990463256836, 9.117092645283265e-07],\n",
       " [0.9999983310699463, 1.6939279703365173e-06],\n",
       " [0.9999977350234985, 2.2119008917798055e-06],\n",
       " [0.9999990463256836, 8.976224421530787e-07],\n",
       " [0.9999979734420776, 1.970954372154665e-06],\n",
       " [0.9999990463256836, 1.006290972327406e-06],\n",
       " [0.9999992847442627, 7.00414545917738e-07],\n",
       " [0.9999992847442627, 7.621592317264003e-07],\n",
       " [0.9999991655349731, 8.607229915469361e-07],\n",
       " [0.9999983310699463, 1.6534136193513405e-06],\n",
       " [0.9999459981918335, 5.3968760767020285e-05],\n",
       " [0.9999972581863403, 2.734414920269046e-06],\n",
       " [0.9999991655349731, 8.569095939492399e-07],\n",
       " [0.9999992847442627, 6.894727562212211e-07],\n",
       " [0.9999934434890747, 6.547233169840183e-06],\n",
       " [0.9999938011169434, 6.254379059100756e-06],\n",
       " [0.9999940395355225, 5.964591764495708e-06],\n",
       " [0.9999992847442627, 7.507267696382769e-07],\n",
       " [0.9999923706054688, 7.626357273693429e-06],\n",
       " [0.9999873638153076, 1.2634640370379202e-05],\n",
       " [0.9999938011169434, 6.222762749530375e-06],\n",
       " [0.9999991655349731, 8.013812475837767e-07],\n",
       " [0.9999967813491821, 3.206229848728981e-06],\n",
       " [0.9999984502792358, 1.533348836346704e-06],\n",
       " [0.9999638795852661, 3.6166966310702264e-05],\n",
       " [0.9999923706054688, 7.655220542801544e-06],\n",
       " [0.9999992847442627, 6.891638122397126e-07],\n",
       " [0.9999992847442627, 7.072772518768033e-07],\n",
       " [0.9999991655349731, 8.926550663090893e-07],\n",
       " [0.9999992847442627, 7.220136808427924e-07],\n",
       " [0.9999991655349731, 8.087248488664045e-07],\n",
       " [0.9999990463256836, 9.473678801441565e-07],\n",
       " [0.9999945163726807, 5.4993925004964694e-06],\n",
       " [0.9998520612716675, 0.00014786073006689548],\n",
       " [0.9999053478240967, 9.468013740843162e-05],\n",
       " [0.9999985694885254, 1.4185827694745967e-06],\n",
       " [0.9999969005584717, 3.0997589419712313e-06],\n",
       " [0.9999992847442627, 7.226722686937137e-07],\n",
       " [0.9999986886978149, 1.3444225714920321e-06],\n",
       " [0.9999390840530396, 6.087297879275866e-05],\n",
       " [0.999969482421875, 3.053057662327774e-05],\n",
       " [0.9999992847442627, 7.096473950696236e-07],\n",
       " [0.9999923706054688, 7.67535311751999e-06],\n",
       " [0.9999048709869385, 9.5160590717569e-05],\n",
       " [0.9999936819076538, 6.2959279603092e-06],\n",
       " [0.9999953508377075, 4.665314918383956e-06],\n",
       " [0.9999985694885254, 1.3766012898486224e-06],\n",
       " [0.9999939203262329, 6.1086693676770665e-06],\n",
       " [0.9999991655349731, 8.327954219566891e-07],\n",
       " [0.9999748468399048, 2.512367609597277e-05],\n",
       " [0.9999938011169434, 6.178383500810014e-06],\n",
       " [0.9999992847442627, 7.43722580409667e-07],\n",
       " [1.0, 0.0],\n",
       " [0.9999974966049194, 2.5561921574990265e-06],\n",
       " [0.9999990463256836, 9.056501539816963e-07],\n",
       " [0.9999979734420776, 2.037893409578828e-06],\n",
       " [0.9999630451202393, 3.696974090416916e-05],\n",
       " [0.9999932050704956, 6.778411261620931e-06],\n",
       " [0.999956488609314, 4.348675429355353e-05],\n",
       " [0.9999980926513672, 1.940327820193488e-06],\n",
       " [0.9999990463256836, 9.465225616622774e-07],\n",
       " [0.9999986886978149, 1.3469327768689254e-06],\n",
       " [0.9999992847442627, 6.883007586111489e-07],\n",
       " [0.9999754428863525, 2.4528082576580346e-05],\n",
       " [0.9999938011169434, 6.169833795865998e-06],\n",
       " [0.9999988079071045, 1.1813795026682783e-06],\n",
       " [0.999998927116394, 1.0424296306155156e-06],\n",
       " [0.9999992847442627, 7.241976618388435e-07],\n",
       " [0.9999992847442627, 6.976320037210826e-07],\n",
       " [0.999993085861206, 6.949572707526386e-06],\n",
       " [0.9999864101409912, 1.3557757483795285e-05],\n",
       " [0.9996967315673828, 0.00030331386369653046],\n",
       " [0.9999160766601562, 8.39193380670622e-05],\n",
       " [0.9999432563781738, 5.676120053976774e-05],\n",
       " [0.9999266862869263, 7.327796629397199e-05],\n",
       " [0.9999637603759766, 3.6206129152560607e-05],\n",
       " [0.9999992847442627, 7.667426871194039e-07],\n",
       " [0.9999992847442627, 6.848723046459781e-07],\n",
       " [0.9999992847442627, 6.866747526146355e-07],\n",
       " [0.9999936819076538, 6.268608103709994e-06],\n",
       " [0.9999990463256836, 9.262132039111748e-07],\n",
       " [0.9999963045120239, 3.719673031810089e-06],\n",
       " [0.9999990463256836, 9.655371968619875e-07],\n",
       " [0.9999984502792358, 1.5652550473532756e-06],\n",
       " [0.9999992847442627, 6.981665023886308e-07],\n",
       " [0.9999977350234985, 2.2071428702474805e-06],\n",
       " [0.9999991655349731, 8.536681548321212e-07],\n",
       " [0.9999992847442627, 7.22946595033136e-07],\n",
       " [0.9999992847442627, 7.661930112590198e-07],\n",
       " [0.9999991655349731, 8.157816182574607e-07],\n",
       " [0.9999991655349731, 7.758118272249703e-07],\n",
       " [0.9999954700469971, 4.540957434073789e-06],\n",
       " [0.9999991655349731, 8.208425015254761e-07],\n",
       " [0.999998927116394, 1.0536667787164333e-06],\n",
       " [0.9999985694885254, 1.385717382618168e-06],\n",
       " [0.9999991655349731, 8.860175739755505e-07],\n",
       " [0.9999992847442627, 6.903748612785421e-07],\n",
       " [0.9999972581863403, 2.760072220553411e-06],\n",
       " [0.999998927116394, 1.1141534059788682e-06],\n",
       " [0.9999986886978149, 1.2733399898934294e-06],\n",
       " [0.9999991655349731, 8.623136409369181e-07],\n",
       " [0.9999990463256836, 9.661249578130082e-07],\n",
       " [0.9999992847442627, 7.715942729191738e-07],\n",
       " [0.999998927116394, 1.0452816923134378e-06],\n",
       " [0.9999938011169434, 6.154472430353053e-06],\n",
       " [0.9999960660934448, 3.939735051972093e-06],\n",
       " [0.9999985694885254, 1.3901236570745823e-06],\n",
       " [0.9999972581863403, 2.743717004705104e-06],\n",
       " [0.9999992847442627, 7.713323384450632e-07],\n",
       " [0.9999992847442627, 7.728233981652011e-07],\n",
       " [0.9999992847442627, 7.598230808980588e-07],\n",
       " [0.9999991655349731, 8.053767714955029e-07],\n",
       " [0.9999991655349731, 8.074694051174447e-07],\n",
       " [0.9999988079071045, 1.178130105472519e-06],\n",
       " [0.999998927116394, 1.0957613767459407e-06],\n",
       " [0.9999991655349731, 8.20142986412975e-07],\n",
       " [0.9999990463256836, 9.040502959578589e-07],\n",
       " [0.9999990463256836, 9.114172030422196e-07],\n",
       " [0.9999984502792358, 1.5843160099393572e-06],\n",
       " [0.9999991655349731, 8.108957558761176e-07],\n",
       " [0.9999936819076538, 6.263856903387932e-06],\n",
       " [0.9999992847442627, 7.747915446998377e-07],\n",
       " [0.9999985694885254, 1.4404330386241782e-06],\n",
       " [0.9999984502792358, 1.5247003375407076e-06],\n",
       " [0.9999988079071045, 1.144853172263538e-06],\n",
       " [0.999997615814209, 2.3876739305705996e-06],\n",
       " [0.9999487400054932, 5.1224858907517046e-05],\n",
       " [0.9999502897262573, 4.971923044649884e-05],\n",
       " [0.9999988079071045, 1.2359562333585927e-06],\n",
       " [0.9999991655349731, 8.406465212829062e-07],\n",
       " [0.9999665021896362, 3.352836210979149e-05],\n",
       " [0.9999924898147583, 7.466714578185929e-06],\n",
       " [0.9999983310699463, 1.6509555962329614e-06],\n",
       " [0.9999991655349731, 8.845021852721402e-07],\n",
       " [0.9999991655349731, 8.429473155047162e-07],\n",
       " [0.9999985694885254, 1.4640127119491808e-06],\n",
       " [0.9999990463256836, 8.961256412476359e-07],\n",
       " [0.9999951124191284, 4.8755318857729435e-06],\n",
       " [0.9999963045120239, 3.694328825076809e-06],\n",
       " [0.9999991655349731, 8.334095582540613e-07],\n",
       " [0.9999971389770508, 2.8684062272077426e-06],\n",
       " [0.999998927116394, 1.0887863481912063e-06],\n",
       " [0.9999985694885254, 1.4679580999654718e-06],\n",
       " [0.9999990463256836, 9.481198617322661e-07],\n",
       " [0.9999990463256836, 1.0024605217040516e-06],\n",
       " [0.999998927116394, 1.0839521564776078e-06],\n",
       " [0.9999862909317017, 1.3709278391615953e-05],\n",
       " [0.9999934434890747, 6.5362532950530294e-06],\n",
       " [0.9999988079071045, 1.1743510412998148e-06],\n",
       " [0.9999947547912598, 5.188247541809687e-06],\n",
       " [0.9999932050704956, 6.75454111842555e-06],\n",
       " [0.9999990463256836, 9.818709258979652e-07],\n",
       " [0.9999992847442627, 7.055694482005492e-07],\n",
       " [0.9999939203262329, 6.100431619415758e-06],\n",
       " [0.9999992847442627, 7.63947241466667e-07],\n",
       " [0.9999991655349731, 8.897628163140325e-07],\n",
       " [0.9999992847442627, 7.382431590485794e-07],\n",
       " [0.9999922513961792, 7.791605639795307e-06],\n",
       " [0.99996018409729, 3.98641059291549e-05],\n",
       " [0.9999821186065674, 1.7867921997094527e-05],\n",
       " [0.9999971389770508, 2.8272311283217277e-06],\n",
       " [0.9999991655349731, 8.490596883348189e-07],\n",
       " [0.9999991655349731, 8.39896415527619e-07],\n",
       " [0.9999991655349731, 8.373012292395288e-07],\n",
       " [0.9999990463256836, 9.229243573827262e-07],\n",
       " [0.9999992847442627, 7.698706667724764e-07],\n",
       " [0.9999936819076538, 6.29416308584041e-06],\n",
       " [0.9999945163726807, 5.472735665534856e-06],\n",
       " [0.9999992847442627, 6.875397389194404e-07],\n",
       " [0.999998927116394, 1.0170753057536785e-06],\n",
       " [0.9999990463256836, 9.631571629142854e-07],\n",
       " [0.9999991655349731, 8.200491379284358e-07],\n",
       " [0.9999990463256836, 9.228856470144819e-07],\n",
       " [0.9999936819076538, 6.262775514187524e-06],\n",
       " [0.999997615814209, 2.333636302864761e-06],\n",
       " [0.9999992847442627, 7.553090881629032e-07],\n",
       " [0.9999992847442627, 7.626907745361677e-07],\n",
       " [0.9999990463256836, 9.550823278914322e-07],\n",
       " [0.9999992847442627, 7.172795335463888e-07],\n",
       " [0.9999992847442627, 7.744473009552166e-07],\n",
       " [0.9999938011169434, 6.177546765684383e-06],\n",
       " [0.9999122619628906, 8.768308180151507e-05],\n",
       " [0.9999837875366211, 1.623764546820894e-05],\n",
       " [0.9999990463256836, 9.189984098156856e-07],\n",
       " [0.9999681711196899, 3.183413718943484e-05],\n",
       " [0.9999912977218628, 8.661224455863703e-06],\n",
       " [0.9999809265136719, 1.9013970813830383e-05],\n",
       " [0.9994217157363892, 0.0005783113301731646],\n",
       " [0.9998703002929688, 0.0001297177659580484],\n",
       " [0.9999992847442627, 7.341829473261896e-07],\n",
       " [0.9999991655349731, 8.084086857707007e-07],\n",
       " [0.9999938011169434, 6.192907221702626e-06],\n",
       " [0.9999470710754395, 5.298294126987457e-05],\n",
       " [0.9999363422393799, 6.361377745633945e-05],\n",
       " [0.9999771118164062, 2.2841955797048286e-05],\n",
       " [0.9999992847442627, 7.024186743365135e-07],\n",
       " [0.9999980926513672, 1.8975973716806038e-06],\n",
       " [0.9999936819076538, 6.32515821052948e-06],\n",
       " [0.9999901056289673, 9.856678843789268e-06],\n",
       " [0.9999352693557739, 6.469168874900788e-05],\n",
       " [1.0, 0.0],\n",
       " [1.0, 0.0],\n",
       " [1.0, 0.0],\n",
       " [0.9999988079071045, 1.1353070021868916e-06],\n",
       " [0.9999725818634033, 2.7476096875034273e-05],\n",
       " [0.9999991655349731, 8.499281989315932e-07],\n",
       " [0.9999991655349731, 8.464854772682884e-07],\n",
       " [0.9999991655349731, 8.075541018115473e-07],\n",
       " [0.9999986886978149, 1.3449330253934022e-06],\n",
       " [0.9999986886978149, 1.2856947932959883e-06],\n",
       " [0.9999992847442627, 7.373116091002885e-07],\n",
       " [0.9999934434890747, 6.597098945348989e-06],\n",
       " [0.9999986886978149, 1.3111267662679893e-06],\n",
       " [0.9999991655349731, 8.691480957168096e-07],\n",
       " [0.9999990463256836, 9.399344094163098e-07],\n",
       " [0.9999985694885254, 1.3799069620290538e-06],\n",
       " [0.9999991655349731, 8.112004934446304e-07],\n",
       " [0.9999992847442627, 7.256467142724432e-07],\n",
       " [0.9999936819076538, 6.289548764470965e-06],\n",
       " [0.999901294708252, 9.86602099146694e-05],\n",
       " [0.9999939203262329, 6.131332611403195e-06],\n",
       " [0.9999939203262329, 6.111956281529274e-06],\n",
       " [0.9999991655349731, 8.095991006484837e-07],\n",
       " [0.999994158744812, 5.8844357226917055e-06],\n",
       " [0.9999992847442627, 7.293154453691386e-07],\n",
       " [0.9999991655349731, 7.893460747254721e-07],\n",
       " [0.9999992847442627, 7.334096494560072e-07],\n",
       " [0.9999991655349731, 8.935911637308891e-07],\n",
       " [0.9999991655349731, 7.862679467507405e-07],\n",
       " [0.999998927116394, 1.116981138693518e-06],\n",
       " [0.9999992847442627, 7.551203680122853e-07],\n",
       " [0.9999992847442627, 7.274468316609273e-07],\n",
       " [0.9999991655349731, 7.957729621921317e-07],\n",
       " [0.9999990463256836, 9.030765681927733e-07],\n",
       " [0.9999990463256836, 9.842447070695926e-07],\n",
       " [0.9999990463256836, 9.6140252026089e-07],\n",
       " [0.9999988079071045, 1.2411901479936205e-06],\n",
       " [0.9999990463256836, 9.059542094291828e-07],\n",
       " [0.9999946355819702, 5.3897906582278665e-06],\n",
       " [0.9999939203262329, 6.0889351516379975e-06],\n",
       " [0.999995231628418, 4.762249773193616e-06],\n",
       " [0.9999992847442627, 7.286598133759981e-07],\n",
       " [0.9999988079071045, 1.1790450571425026e-06],\n",
       " [0.9999988079071045, 1.1692428643073072e-06],\n",
       " [0.9999986886978149, 1.3221385870565427e-06],\n",
       " [0.999998927116394, 1.0586435337245348e-06],\n",
       " [0.9999990463256836, 9.180225788441021e-07],\n",
       " [0.9999990463256836, 9.588295597495744e-07],\n",
       " [0.9999990463256836, 9.257513511329307e-07],\n",
       " [0.9999940395355225, 5.921974207012681e-06],\n",
       " [0.9999991655349731, 8.817951879791508e-07],\n",
       " [0.9999963045120239, 3.686997160912142e-06],\n",
       " [0.9999938011169434, 6.186614427861059e-06],\n",
       " [0.999992847442627, 7.191520126070827e-06],\n",
       " [0.9999955892562866, 4.422355686983792e-06],\n",
       " [0.9999940395355225, 6.003253929520724e-06],\n",
       " [0.9999992847442627, 7.720624353169114e-07],\n",
       " [0.9999939203262329, 6.1075916164554656e-06],\n",
       " [0.9999938011169434, 6.142375696072122e-06],\n",
       " [0.9999990463256836, 9.329863246421155e-07],\n",
       " [0.9999988079071045, 1.152904019363632e-06],\n",
       " [0.9999986886978149, 1.3364229971557506e-06],\n",
       " [0.9999985694885254, 1.459273335058242e-06],\n",
       " [0.999998927116394, 1.0440045343784732e-06],\n",
       " [0.9999988079071045, 1.1454341120042955e-06],\n",
       " [0.9999985694885254, 1.4653340940640192e-06],\n",
       " [0.9999986886978149, 1.3498363387043355e-06],\n",
       " [0.9999985694885254, 1.4665783965028822e-06],\n",
       " [0.999998927116394, 1.0617910675136955e-06],\n",
       " [0.9999984502792358, 1.5039436220831703e-06],\n",
       " [0.9999985694885254, 1.4813773532296182e-06],\n",
       " [0.9999985694885254, 1.4724614629813004e-06],\n",
       " [0.9999985694885254, 1.3771684734820155e-06],\n",
       " [0.9999985694885254, 1.4675856618850958e-06],\n",
       " [0.9999954700469971, 4.563167294691084e-06],\n",
       " [0.9999991655349731, 8.187456614905386e-07],\n",
       " [0.9999988079071045, 1.2394524446790456e-06],\n",
       " [0.9999991655349731, 8.104017297227983e-07],\n",
       " [0.9999890327453613, 1.0942128938040696e-05],\n",
       " [0.9998582601547241, 0.00014174790703691542],\n",
       " [0.9999710321426392, 2.8974756787647493e-05],\n",
       " [0.9999991655349731, 8.285832677756844e-07],\n",
       " [0.9999991655349731, 8.584399324718106e-07],\n",
       " [0.9999939203262329, 6.092617695685476e-06],\n",
       " [0.9999992847442627, 6.876380780340696e-07],\n",
       " [0.999998927116394, 1.0545132909101085e-06],\n",
       " [0.9999991655349731, 7.813800380063185e-07],\n",
       " [0.9999992847442627, 7.555663046332484e-07],\n",
       " [0.9999938011169434, 6.2298463490151335e-06],\n",
       " [0.9999992847442627, 6.852139904367505e-07],\n",
       " [0.9999990463256836, 9.241414886673738e-07],\n",
       " [0.9999988079071045, 1.2173005643489887e-06],\n",
       " [0.9999991655349731, 8.087225182862312e-07],\n",
       " [0.9999990463256836, 1.0117644251295133e-06],\n",
       " [0.9999990463256836, 9.125618021244009e-07],\n",
       " [0.9999973773956299, 2.617431619000854e-06],\n",
       " [0.9999934434890747, 6.520147053379333e-06],\n",
       " [0.9998410940170288, 0.00015889311907812953],\n",
       " [0.999995231628418, 4.719407570519252e-06],\n",
       " [0.9999798536300659, 2.0180417777737603e-05],\n",
       " [0.9999992847442627, 7.37809614292928e-07],\n",
       " [0.9999520778656006, 4.79213340440765e-05],\n",
       " [0.9999872446060181, 1.2806915947294328e-05],\n",
       " [0.9999992847442627, 6.991019176894042e-07],\n",
       " [0.9999829530715942, 1.7016536730807275e-05],\n",
       " [0.9999643564224243, 3.565016595530324e-05],\n",
       " [0.9999939203262329, 6.1276382439245936e-06],\n",
       " [0.9990540146827698, 0.0009459839784540236],\n",
       " [0.999862790107727, 0.00013723447045776993],\n",
       " [0.9999992847442627, 7.229211291814863e-07],\n",
       " [0.9999992847442627, 7.518631264247233e-07],\n",
       " [0.9999927282333374, 7.286644176929258e-06],\n",
       " [0.9999988079071045, 1.1915163895537262e-06],\n",
       " [0.9999938011169434, 6.255351308936952e-06],\n",
       " [0.9999990463256836, 9.453281677451741e-07],\n",
       " [0.9999992847442627, 7.54831773974729e-07],\n",
       " [0.999998927116394, 1.0224280231341254e-06],\n",
       " [0.999998927116394, 1.050337004926405e-06],\n",
       " [0.9999957084655762, 4.27480335929431e-06],\n",
       " [0.9999915361404419, 8.460693607048597e-06],\n",
       " [0.9999938011169434, 6.247493274713634e-06],\n",
       " [0.9999942779541016, 5.680874437530292e-06],\n",
       " [0.9999939203262329, 6.101129656599369e-06],\n",
       " [0.9999945163726807, 5.503175088961143e-06],\n",
       " [0.9999922513961792, 7.790319614287e-06],\n",
       " [0.9999982118606567, 1.7769334590411745e-06],\n",
       " [0.9999951124191284, 4.890172021987382e-06],\n",
       " [0.9999992847442627, 7.74417742377409e-07],\n",
       " [0.9999954700469971, 4.531721060629934e-06],\n",
       " [0.9999938011169434, 6.256436790863518e-06],\n",
       " [0.9999387264251709, 6.121777551015839e-05],\n",
       " [0.99979168176651, 0.00020836153998970985],\n",
       " [0.9996082186698914, 0.0003917556023225188],\n",
       " [0.9999926090240479, 7.3445221460133325e-06],\n",
       " [0.999997615814209, 2.342439074709546e-06],\n",
       " [0.9998698234558105, 0.00013019348261877894],\n",
       " [0.9999228715896606, 7.710690988460556e-05],\n",
       " [0.9999843835830688, 1.5589337635901757e-05],\n",
       " [0.9999992847442627, 6.992519274717779e-07],\n",
       " [0.9999938011169434, 6.243574262043694e-06],\n",
       " [0.9999833106994629, 1.6720865460229106e-05],\n",
       " [0.9999781847000122, 2.1868654584977776e-05],\n",
       " [0.9999992847442627, 6.952940339033375e-07],\n",
       " [1.0, 0.0],\n",
       " [0.9999984502792358, 1.498494725638011e-06],\n",
       " [1.0, 0.0],\n",
       " [0.9999992847442627, 7.216743256321934e-07],\n",
       " [0.9999990463256836, 9.229050306203135e-07],\n",
       " [0.9999992847442627, 7.001534072514914e-07],\n",
       " [0.9999983310699463, 1.6549296333323582e-06],\n",
       " [0.9999990463256836, 9.138690302279429e-07],\n",
       " [0.9999992847442627, 7.398346610898443e-07],\n",
       " [0.9999990463256836, 9.177538231597282e-07],\n",
       " [0.9999982118606567, 1.7598798649487435e-06],\n",
       " [0.9999992847442627, 7.555756837973604e-07],\n",
       " [0.9999978542327881, 2.0969116576452507e-06],\n",
       " [0.9999990463256836, 9.128438023253693e-07],\n",
       " [0.9999991655349731, 8.398483828386816e-07],\n",
       " [0.9999992847442627, 7.04687863617437e-07],\n",
       " [0.9999990463256836, 1.012606162476004e-06],\n",
       " [0.9999983310699463, 1.7045016420524917e-06],\n",
       " [0.9999990463256836, 9.184078635371407e-07],\n",
       " [0.9999991655349731, 7.8845300777175e-07],\n",
       " [0.9999985694885254, 1.4420508023249567e-06],\n",
       " [0.9999967813491821, 3.253231625421904e-06],\n",
       " [0.9999991655349731, 8.151742463269329e-07],\n",
       " [0.9999594688415527, 4.047798574902117e-05],\n",
       " [0.9999791383743286, 2.0876592316199094e-05],\n",
       " [0.9999992847442627, 6.897345201650751e-07],\n",
       " [0.9999991655349731, 8.496623422615812e-07],\n",
       " [0.9999992847442627, 7.022914019216842e-07],\n",
       " [0.9999992847442627, 6.912114827173355e-07],\n",
       " [0.9999938011169434, 6.203593784448458e-06],\n",
       " [0.9999938011169434, 6.214939730853075e-06],\n",
       " [0.9999939203262329, 6.070161362004001e-06],\n",
       " [0.9999991655349731, 8.140571026160615e-07],\n",
       " [0.9999938011169434, 6.251558261283208e-06],\n",
       " [0.9998902082443237, 0.00010980304068652913],\n",
       " [0.9999939203262329, 6.102794031903613e-06],\n",
       " [0.9999991655349731, 7.783166324770718e-07],\n",
       " [0.9999992847442627, 7.702106472606829e-07],\n",
       " [0.9999983310699463, 1.7064729718185845e-06],\n",
       " [0.9999992847442627, 7.71654640630004e-07],\n",
       " [0.9999979734420776, 2.078015768347541e-06],\n",
       " [0.9999938011169434, 6.249066700547701e-06],\n",
       " [0.9999992847442627, 7.522525606873387e-07],\n",
       " [0.9999991655349731, 7.917487891972996e-07],\n",
       " [0.9999936819076538, 6.299170763668371e-06],\n",
       " [0.9999938011169434, 6.204694727784954e-06],\n",
       " [0.9999992847442627, 7.572376716780127e-07],\n",
       " [0.9999991655349731, 8.253025498561328e-07],\n",
       " [0.9999991655349731, 7.863017117415438e-07],\n",
       " [0.999998927116394, 1.1058057225454831e-06],\n",
       " [0.999998927116394, 1.0468229447724298e-06],\n",
       " [0.9999992847442627, 7.149636758185807e-07],\n",
       " [0.999998927116394, 1.0186109875576221e-06],\n",
       " [0.9999991655349731, 7.971569289111358e-07],\n",
       " [0.9999936819076538, 6.282085450948216e-06],\n",
       " [0.9999959468841553, 4.10149959861883e-06],\n",
       " [0.9999990463256836, 9.193543064611731e-07],\n",
       " [0.9999991655349731, 7.875466963014333e-07],\n",
       " [0.9999938011169434, 6.139260221971199e-06],\n",
       " [0.9999992847442627, 6.994346790634154e-07],\n",
       " [0.9999932050704956, 6.757427854608977e-06],\n",
       " [0.9999885559082031, 1.1470036042737775e-05],\n",
       " [0.9993405938148499, 0.0006593946018256247],\n",
       " [0.999993085861206, 6.950235729163978e-06],\n",
       " [0.999998927116394, 1.109572735913389e-06],\n",
       " [0.9999990463256836, 9.779309948498849e-07],\n",
       " [0.999998927116394, 1.1320938710923656e-06],\n",
       " [0.9999986886978149, 1.3628892929773428e-06],\n",
       " [0.999998927116394, 1.014826580103545e-06],\n",
       " [0.9999986886978149, 1.2820278243452776e-06],\n",
       " [0.9999985694885254, 1.4157823216010001e-06],\n",
       " [0.9999985694885254, 1.4600834674638463e-06],\n",
       " [0.9999992847442627, 7.338251748478797e-07],\n",
       " [0.9999992847442627, 7.002582833592896e-07],\n",
       " [0.999998927116394, 1.118772274821822e-06],\n",
       " [0.9999988079071045, 1.1933163932553725e-06],\n",
       " [0.9999986886978149, 1.3590280332209659e-06],\n",
       " [0.9999984502792358, 1.5001434121586499e-06],\n",
       " [0.9999986886978149, 1.2830797686547157e-06],\n",
       " [0.9999990463256836, 9.876371223072056e-07],\n",
       " [0.999998927116394, 1.105240585275169e-06],\n",
       " [0.999998927116394, 1.1171622418260085e-06],\n",
       " [0.9999992847442627, 7.049983992146736e-07],\n",
       " [0.9999938011169434, 6.162400950415758e-06],\n",
       " [0.9999938011169434, 6.166510502225719e-06],\n",
       " [0.999983549118042, 1.6496493117301725e-05],\n",
       " [0.9999992847442627, 7.26271196072048e-07],\n",
       " [0.9999992847442627, 7.377673796327144e-07],\n",
       " [0.9999936819076538, 6.266958280320978e-06],\n",
       " [0.9999974966049194, 2.5475380880379817e-06],\n",
       " [0.9999939203262329, 6.100088285165839e-06],\n",
       " [0.9999991655349731, 8.276260814454872e-07],\n",
       " [0.9999990463256836, 1.0022961305367062e-06],\n",
       " [0.9999970197677612, 3.019743189724977e-06],\n",
       " [0.9999992847442627, 7.274107360899507e-07],\n",
       " [0.999998927116394, 1.1276619034106261e-06],\n",
       " [0.9999990463256836, 9.849602520262124e-07],\n",
       " [0.9999991655349731, 8.216821356654691e-07],\n",
       " [0.9999988079071045, 1.2251806538188248e-06],\n",
       " [0.9999991655349731, 7.848235554774874e-07],\n",
       " [0.9999991655349731, 8.067797239164065e-07],\n",
       " [0.9999992847442627, 7.445131018357642e-07],\n",
       " [0.9999973773956299, 2.654912805155618e-06],\n",
       " [0.9999961853027344, 3.7805052670591976e-06],\n",
       " [0.9999361038208008, 6.392404611688107e-05],\n",
       " [0.9998708963394165, 0.0001291101216338575],\n",
       " [0.9999936819076538, 6.309981472440995e-06],\n",
       " [0.9999991655349731, 7.773508059472078e-07],\n",
       " [0.9999983310699463, 1.696927256489289e-06],\n",
       " [0.9999966621398926, 3.2880684557312634e-06],\n",
       " [0.9999973773956299, 2.582096840342274e-06],\n",
       " [0.9999959468841553, 4.019006155431271e-06],\n",
       " [0.9999973773956299, 2.571841150711407e-06],\n",
       " [0.9999936819076538, 6.2807616814097855e-06],\n",
       " [0.9999595880508423, 4.035207166452892e-05],\n",
       " [0.9999939203262329, 6.13189422438154e-06],\n",
       " [0.9999991655349731, 8.640728310638224e-07],\n",
       " [0.9999990463256836, 9.544230579194846e-07],\n",
       " [0.9999939203262329, 6.1123582781874575e-06],\n",
       " [0.999998927116394, 1.1014966503353207e-06],\n",
       " [0.9999992847442627, 7.370501293735288e-07],\n",
       " [0.999998927116394, 1.025914912133885e-06],\n",
       " [0.9999990463256836, 9.463357741879008e-07],\n",
       " [0.9999955892562866, 4.416863703227136e-06],\n",
       " [0.9999914169311523, 8.579361747251824e-06],\n",
       " [0.9999938011169434, 6.235945420485223e-06],\n",
       " [0.999995231628418, 4.760665433423128e-06],\n",
       " [0.9999992847442627, 7.133141366466589e-07],\n",
       " [0.9998831748962402, 0.0001167662485386245],\n",
       " [0.9999972581863403, 2.7592352580541046e-06],\n",
       " [0.9999992847442627, 7.00359805705375e-07],\n",
       " [0.9999936819076538, 6.2765520851826295e-06],\n",
       " [0.999995231628418, 4.799500402441481e-06],\n",
       " [0.9999992847442627, 7.228431968542282e-07],\n",
       " [0.9999946355819702, 5.345430054148892e-06],\n",
       " [0.9998559951782227, 0.00014394958270713687],\n",
       " [0.9999631643295288, 3.682667011162266e-05],\n",
       " [0.999665379524231, 0.00033460636041127145],\n",
       " [0.9999991655349731, 8.224504313147918e-07],\n",
       " [0.9999991655349731, 7.874265293139615e-07],\n",
       " [0.999996542930603, 3.4605038763402263e-06],\n",
       " [0.999897837638855, 0.00010214740177616477],\n",
       " [0.9996019005775452, 0.00039813833427615464],\n",
       " [0.9998194575309753, 0.00018052099039778113],\n",
       " [0.9999971389770508, 2.8280537662794814e-06],\n",
       " [0.9999990463256836, 9.60033617047884e-07],\n",
       " [0.9999992847442627, 7.648030191376165e-07],\n",
       " [0.999998927116394, 1.051021399689489e-06],\n",
       " [0.9999991655349731, 8.217965614676359e-07],\n",
       " [0.9999992847442627, 6.896845547998964e-07],\n",
       " [0.9999992847442627, 7.540266437899845e-07],\n",
       " [0.9999986886978149, 1.3404923038251582e-06],\n",
       " [0.9999990463256836, 9.533696356811561e-07],\n",
       " [0.9999903440475464, 9.597017196938396e-06],\n",
       " [0.9986391663551331, 0.0013608820736408234],\n",
       " [0.9999734163284302, 2.658367156982422e-05],\n",
       " [0.9999992847442627, 7.201507514764671e-07],\n",
       " [0.999998927116394, 1.062946125784947e-06],\n",
       " [0.9999990463256836, 9.212342320097378e-07],\n",
       " [0.9999980926513672, 1.903339125419734e-06],\n",
       " [0.9999939203262329, 6.099285656091524e-06],\n",
       " [0.9999992847442627, 7.284784828698321e-07],\n",
       " [0.9999992847442627, 7.746903065708466e-07],\n",
       " [0.9999939203262329, 6.103387931943871e-06],\n",
       " [0.9999992847442627, 7.031894142528472e-07],\n",
       " [0.9999991655349731, 8.663625976623734e-07],\n",
       " [0.9999992847442627, 7.03622788478242e-07],\n",
       " [0.9999895095825195, 1.0486931387276854e-05],\n",
       " [0.9998834133148193, 0.00011658112634904683],\n",
       " [0.9999939203262329, 6.110691174399108e-06],\n",
       " [0.9999992847442627, 6.957782829886128e-07],\n",
       " [0.9999992847442627, 7.103062102942204e-07],\n",
       " [0.9999938011169434, 6.193787157826591e-06],\n",
       " [0.9999992847442627, 7.050763883853506e-07],\n",
       " [0.9999992847442627, 7.19281786132342e-07],\n",
       " [0.9999992847442627, 7.726664534857264e-07],\n",
       " [0.9999992847442627, 7.698170634284907e-07],\n",
       " [0.9999969005584717, 3.070741968258517e-06],\n",
       " [0.999984860420227, 1.5098865333129652e-05],\n",
       " [0.9999938011169434, 6.191938609845238e-06],\n",
       " [0.9999992847442627, 7.020362886578368e-07],\n",
       " [0.9999991655349731, 7.971630111569539e-07],\n",
       " [0.9999949932098389, 5.062609034212073e-06],\n",
       " [0.9999992847442627, 7.738581189187244e-07],\n",
       " [0.9999992847442627, 7.442326932505239e-07],\n",
       " [0.9999990463256836, 9.112172847380862e-07],\n",
       " [0.999998927116394, 1.0423182175145485e-06],\n",
       " [0.9999990463256836, 9.864342018772732e-07],\n",
       " [0.9999992847442627, 7.489840072594234e-07],\n",
       " [0.9999980926513672, 1.8912668338089134e-06],\n",
       " [0.9999970197677612, 3.0267581223597517e-06],\n",
       " [0.9999939203262329, 6.117496013757773e-06],\n",
       " [0.9999837875366211, 1.6260455595329404e-05],\n",
       " [0.9999992847442627, 6.851754505987628e-07],\n",
       " [0.9999986886978149, 1.320530714110646e-06],\n",
       " [0.9999991655349731, 8.315549848703085e-07],\n",
       " [0.9999986886978149, 1.2525483725767117e-06],\n",
       " [0.9999986886978149, 1.256114273928688e-06],\n",
       " [0.9999990463256836, 1.0044270766229602e-06],\n",
       " [0.9999990463256836, 9.554840971759404e-07],\n",
       " [0.9999990463256836, 9.59425960900262e-07],\n",
       " ...]"
      ]
     },
     "execution_count": 218,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "predictions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 123,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[]"
      ]
     },
     "execution_count": 123,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_scores"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 118,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0., 1.],\n",
       "       [0., 1.],\n",
       "       [0., 1.],\n",
       "       ...,\n",
       "       [1., 0.],\n",
       "       [1., 0.],\n",
       "       [1., 0.]], dtype=float32)"
      ]
     },
     "execution_count": 118,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Y"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(40, 40, 3)"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "samp_img.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(50, 50, 3)"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.resize(samp_img, (50, 50, 3)).shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Trying out selective search\n",
    "\n",
    "img = cv2.imread(path.join(base_path, file))\n",
    "image = skimage.io.imread(path.join(base_path, file))\n",
    "boxes = selective_search(image, mode='fast', random=False)\n",
    "predictions = []\n",
    "all_images = []\n",
    "for ix in boxes:\n",
    "    samp_img = img[ix[1]: ix[3], ix[0]: ix[2]]\n",
    "    all_images.append(cv2.resize(samp_img, (40, 40)))\n",
    "all_images = np.array(all_images) * 1.0 / 255\n",
    "predictions.extend(model.predict_proba(all_images, verbose =1).tolist())\n",
    "img = cv2.imread(path.join(base_path, \"labels\", file))\n",
    "all_boxes = []\n",
    "all_scores = []\n",
    "for ix in range(len(predictions)):\n",
    "    prob = predictions[ix][1]\n",
    "    bx = boxes[ix]\n",
    "    new_box = [bx[0], bx[1], bx[2] - bx[0], bx[3] - bx[1]]\n",
    "    all_boxes.append(new_box)\n",
    "    all_scores.append(prob)\n",
    "indices = nms.boxes(all_boxes, all_scores, nms_threshold=0.1)\n",
    "print(len(indices))\n",
    "for ix in tqdm.tqdm_notebook(indices):\n",
    "    if all_scores[ix] > 0.9:\n",
    "        box = boxes[ix]\n",
    "        w = box[2] - box[0]\n",
    "        h = box[3] - box[1]\n",
    "        if w < 80 and h < 80:\n",
    "            img = cv2.rectangle(img, (box[0], box[1]), (box[2], box[3]), (0, 255, 0), 2)\n",
    "cv2.imwrite(path.join(base_path, \"predictions\", file), img)\n",
    "<!-- #     if prob > 0.9:\n",
    "#         box = boxes[ix]\n",
    "#         w = box[2] - box[0]\n",
    "#         h = box[3] - box[1]\n",
    "#         if w < 80 and h < 80:\n",
    "#             img = cv2.rectangle(img, (box[0], box[1]), (box[2], box[3]), (0, 255, 0), 2)\n",
    "# cv2.imwrite(path.join(base_path, \"predictions\", file), img) -->"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "for ix, val in df.iterrows():\n",
    "    img_c = img[val[\"ys\"]:val[\"ye\"], val[\"xs\"]: val[\"xe\"]]\n",
    "    cv2.imwrite(path.join(PATH, \"temp\", \"_\".join(map(str, [ix, val[\"ys\"], val[\"ye\"], val[\"xs\"], val[\"xe\"]])) + \".jpg\"), img_c)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "all_counts = []\n",
    "for ix, val in enumerate(counts):\n",
    "    counts[ix].append(ann[ann.filename==val[0]].iloc[0].region_count)\n",
    "counts = counts[:-3]\n",
    "pd.DataFrame(counts, columns=[\"filename\", \"predicted\", \"actual\"])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "all_cropped = pd.read_csv(path.join(PATH, \"all_cropped_40.csv\"))\n",
    "test_images = [\"2019_1010_123350_243.JPG\", \"2019_1010_123351_244.JPG\"]\n",
    "df_train = all_cropped[~all_cropped[\"filename\"].isin(test_images)]\n",
    "df_test = all_cropped[all_cropped[\"filename\"].isin(test_images)]\n",
    "print(len(df_train), len(all_cropped), len(df_test))\n",
    "\n",
    "df_true = df_train[df_train[\"label\"] == 1]\n",
    "df_false = df_train[df_train[\"label\"] == 0].sample(len(df_true) * 10)\n",
    "<!-- # df_false = df_train[df_train[\"label\"] == 0] -->\n",
    "while len(df_true) < len(df_false):\n",
    "    df_true = df_true.append(df_true, ignore_index=True)\n",
    "df_training = df_true.append(df_false)\n",
    "print(len(df_training))\n",
    "df_true = df_test[df_test[\"label\"] == 1]\n",
    "<!-- # df_false = df_train[df_train[\"label\"] == 0].sample(len(df_true)) -->\n",
    "df_false = df_test[df_test[\"label\"] == 0]\n",
    "df_testing = df_true.append(df_false)\n",
    "df_testing = df_testing.sort_values(\"filename\")\n",
    "print(len(df_testing))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "X = []\n",
    "Y = []\n",
    "file_name = \"\"\n",
    "df_training = df_training.sort_values(\"filename\")\n",
    "for ix, val in df_training.iterrows():\n",
    "    if val[\"filename\"] != file_name:\n",
    "        print(val[\"folder\"], val[\"filename\"])\n",
    "        img = cv2.imread(path.join(PATH, str(val[\"folder\"]), val[\"filename\"]))\n",
    "        file_name = val[\"filename\"]\n",
    "\n",
    "    if img is None:\n",
    "        continue\n",
    "    samp_img = img[val[\"ys\"]: val[\"ye\"], val[\"xs\"]: val[\"xe\"]]\n",
    "    samp_img = cv2.cvtColor(samp_img, cv2.COLOR_BGR2GRAY)\n",
    "    X.append(samp_img)\n",
    "    Y.append(val[\"label\"])\n",
    "X = np.array(X)\n",
    "Y = np.array(Y)\n",
    "Y = to_categorical(Y)\n",
    "print(len(X))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "base_path = path.join(PATH, \"dset_all\", \"train\")\n",
    "for ix, val in enumerate(X):\n",
    "    if Y[ix] == 1:\n",
    "        cv2.imwrite(path.join(base_path, \"positive\", str(ix) + \".jpg\"), val)\n",
    "    else:\n",
    "        cv2.imwrite(path.join(base_path, \"negative\", str(ix) + \".jpg\"), val)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "base_path = path.join(PATH, \"dset_all\", \"test\")\n",
    "for ix, val in enumerate(XT):\n",
    "    if YT[ix] == 1:\n",
    "        cv2.imwrite(path.join(base_path, \"positive\", str(ix) + \".jpg\"), val)\n",
    "    else:\n",
    "        cv2.imwrite(path.join(base_path, \"negative\", str(ix) + \".jpg\"), val)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Create dataset using random range of inputs\n",
    "\n",
    "X = []\n",
    "Y = []\n",
    "folders = [\"1\", \"2\", \"samples\"]\n",
    "test_images = [\"2019_1010_123350_243.JPG\", \"2019_1010_123351_244.JPG\"]\n",
    "for folder in folders:\n",
    "    base_path = path.join(PATH, folder)\n",
    "    print(base_path)\n",
    "    ann = pd.read_csv(path.join(base_path, \"via_export_csv.csv\"))\n",
    "    for file in ann[\"filename\"].unique():\n",
    "        if file in test_images:\n",
    "            continue\n",
    "        cnt1 = 0\n",
    "        cnt0 = 0\n",
    "        print(file)\n",
    "        sample_ann = ann[ann[\"filename\"] == file]\n",
    "        img = cv2.imread(path.join(base_path, file))\n",
    "        if img is None:\n",
    "            continue\n",
    "        # Read all the annotations\n",
    "        for ix, val in sample_ann.iterrows():\n",
    "            coords = json.loads(val[\"region_shape_attributes\"])\n",
    "            if \"x\" not in coords.keys():\n",
    "                continue\n",
    "            x1 = coords[\"x\"]\n",
    "            if x1 is None:\n",
    "                continue\n",
    "            width = coords[\"width\"]\n",
    "            y1 = coords[\"y\"]\n",
    "            height = coords[\"height\"]\n",
    "            width_range = max(1, int((width - 30) / 2))\n",
    "            height_range = max(1, int((height - 30) / 2))\n",
    "            for _ in range(15):\n",
    "                x2 = x1 + width + random.randrange(-width_range, width_range)\n",
    "                y2 = y1 + height + random.randrange(-height_range, height_range)\n",
    "                img_crop = img[y1: y2, x1:x2]\n",
    "                try:\n",
    "                    X.append(cv2.resize(img_crop, (40, 40)))\n",
    "                    Y.append(1)\n",
    "                    cnt1 += 1\n",
    "                except Exception as e:\n",
    "                    print(e)\n",
    "                    pass\n",
    "        print(cnt1)\n",
    "        for ix, val in sample_ann.iterrows():\n",
    "            coords = json.loads(val[\"region_shape_attributes\"])\n",
    "            if \"x\" not in coords.keys():\n",
    "                continue\n",
    "            x1 = coords[\"x\"]\n",
    "            if x1 is None:\n",
    "                continue\n",
    "            x2 = x1 + coords[\"width\"]\n",
    "            y1 = coords[\"y\"]\n",
    "            y2 = y1 + coords[\"height\"]\n",
    "            img_crop = img[y1:y2, x1:x2]\n",
    "            img[y1:y2, x1:x2] = np.ones_like(img_crop)\n",
    "        # img = cv2.imread(path.join(base_path, file))\n",
    "        while not cnt0 == cnt1:\n",
    "            y1 = random.randint(0, 2950)\n",
    "            x1 = random.randint(0, 3950)\n",
    "            y2 = y1 + 40\n",
    "            x2 = x1 + 40\n",
    "            img_crop = img[y1:y2, x1:x2]\n",
    "            if len(np.where(img_crop==1)[0]) < 100:\n",
    "                img_crop = img[y1:y2, x1:x2]\n",
    "                X.append(img_crop.copy())\n",
    "                Y.append(0)\n",
    "                cnt0 += 1\n",
    "                img[y1:y2, x1:x2] = np.ones_like(img_crop)\n",
    "X = np.array(X)\n",
    "Y = np.array(Y)\n",
    "Y = to_categorical(Y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 96,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "/mnt/d/personal/cotton/Annoted_boll_burst_2019_trial/Set1\n",
      "2019_1206_122944_171.JPG\n",
      "399\n",
      "2019_1206_122945_172.JPG\n",
      "773\n",
      "2019_1206_122947_173.JPG\n",
      "733\n",
      "2019_1206_122948_174.JPG\n",
      "910\n",
      "2019_1206_122950_175.JPG\n",
      "1168\n",
      "2019_1206_122951_176.JPG\n",
      "998\n",
      "2019_1206_122953_177.JPG\n",
      "437\n",
      "2019_1206_122954_178.JPG\n",
      "496\n",
      "/mnt/d/personal/cotton/Annoted_boll_burst_2019_trial/Set2\n",
      "2019_1206_122959_181.JPG\n",
      "214\n",
      "2019_1206_123000_182.JPG\n",
      "170\n",
      "2019_1206_123002_183.JPG\n",
      "269\n",
      "2019_1206_123003_184.JPG\n",
      "409\n",
      "2019_1206_123005_185.JPG\n",
      "332\n",
      "2019_1206_123006_186.JPG\n",
      "290\n",
      "2019_1206_123008_187.JPG\n",
      "201\n",
      "2019_1206_123010_188.JPG\n",
      "189\n",
      "2019_1206_123011_189.JPG\n",
      "157\n",
      "2019_1206_123012_190.JPG\n",
      "286\n",
      "/mnt/d/personal/cotton/Annoted_boll_burst_2019_trial/Set3\n",
      "2019_1206_123014_191.JPG\n",
      "142\n",
      "2019_1206_123016_192.JPG\n",
      "202\n",
      "2019_1206_123017_193.JPG\n",
      "208\n",
      "2019_1206_123019_194.JPG\n",
      "150\n",
      "2019_1206_123020_195.JPG\n",
      "119\n",
      "2019_1206_123021_196.JPG\n",
      "267\n",
      "2019_1206_123023_197.JPG\n",
      "234\n",
      "2019_1206_123024_198.JPG\n",
      "206\n",
      "2019_1206_123026_199.JPG\n",
      "167\n",
      "2019_1206_123027_200.JPG\n",
      "145\n",
      "/mnt/d/personal/cotton/Annoted_boll_burst_2019_trial/Set4\n",
      "2019_1206_123029_201.JPG\n",
      "157\n",
      "2019_1206_123035_205.JPG\n",
      "164\n",
      "2019_1206_123036_206.JPG\n",
      "195\n",
      "2019_1206_123038_207.JPG\n",
      "545\n",
      "2019_1206_123039_208.JPG\n",
      "549\n",
      "2019_1206_123041_209.JPG\n",
      "693\n",
      "2019_1206_123042_210.JPG\n",
      "586\n",
      "2019_1206_123044_211.JPG\n",
      "511\n",
      "2019_1206_123045_212.JPG\n",
      "535\n",
      "2019_1206_123047_213.JPG\n",
      "475\n",
      "/mnt/d/personal/cotton/Annoted_boll_burst_2019_trial/Set5\n",
      "2019_1206_123048_214.JPG\n",
      "148\n",
      "2019_1206_123050_215.JPG\n",
      "674\n",
      "2019_1206_123051_216.JPG\n",
      "352\n",
      "2019_1206_123053_217.JPG\n",
      "282\n",
      "2019_1206_123054_218.JPG\n",
      "310\n",
      "2019_1206_123056_219.JPG\n",
      "133\n",
      "2019_1206_123057_220.JPG\n",
      "363\n",
      "2019_1206_123059_221.JPG\n",
      "290\n",
      "2019_1206_123100_222.JPG\n",
      "367\n",
      "2019_1206_123102_223.JPG\n",
      "778\n"
     ]
    }
   ],
   "source": [
    "# Create dataset using the same techique as prediction\n",
    "X = []\n",
    "Y = []\n",
    "folders = [\"Set1\", \"Set2\", \"Set3\", \"Set4\", \"Set5\"]\n",
    "test_images = [\"2019_1206_122956_179.JPG\", \"2019_1206_122957_180.JPG\"]\n",
    "stride = 20\n",
    "width = 40\n",
    "prediction_width = 40\n",
    "do_resize = False\n",
    "for folder in folders:\n",
    "    base_path = path.join(PATH, folder)\n",
    "    print(base_path)\n",
    "    ann = pd.read_csv(path.join(base_path, \"via_export_csv.csv\"))\n",
    "    for file in ann[\"filename\"].unique():\n",
    "        if file in test_images:\n",
    "            continue\n",
    "        cnt1 = 0\n",
    "        cnt0 = 0\n",
    "        print(file)\n",
    "        sample_ann = ann[ann[\"filename\"] == file]\n",
    "        img = cv2.imread(path.join(base_path, file))\n",
    "        if img is None:\n",
    "            continue\n",
    "        img1 = cv2.imread(path.join(base_path, file))\n",
    "        training_data = []\n",
    "\n",
    "        # Create a list of all the boxes\n",
    "        for _x in range(0, img.shape[1] - width, stride):\n",
    "            for _y in range(0, img.shape[0] - width, stride):\n",
    "                xstart = _x\n",
    "                xend = _x + width\n",
    "                ystart = _y\n",
    "                yend = _y + width\n",
    "                training_data.append([ystart, yend, xstart, xend])\n",
    " \n",
    "        # Loop through all annotations and fill the cotton pixels with 1\n",
    "        for ix, val in sample_ann.iterrows():\n",
    "            coords = json.loads(val[\"region_shape_attributes\"])\n",
    "            if \"x\" not in coords.keys():\n",
    "                continue\n",
    "            x1 = coords[\"x\"]\n",
    "            if x1 is None:\n",
    "                continue\n",
    "            x2 = x1 + coords[\"width\"]\n",
    "            y1 = coords[\"y\"]\n",
    "            y2 = y1 + coords[\"height\"]\n",
    "            img_crop = img[y1:y2, x1:x2]\n",
    "            img[y1:y2, x1:x2] = np.ones_like(img_crop) * 200\n",
    " \n",
    "        # Find the ones with the fill less than a value\n",
    "        cotton_ix = []\n",
    "        bg_ix = []      \n",
    "        for ix in range(len(training_data)):\n",
    "            val = training_data[ix]\n",
    "            samp_img = img[val[0]: val[1], val[2]: val[3]]\n",
    "            if len(np.where(samp_img==200)[0]) > 4700:\n",
    "                cotton_ix.append(ix)\n",
    "            else:\n",
    "                bg_ix.append(ix)\n",
    "    \n",
    "        imgg = cv2.imread(path.join(base_path, \"labels\", file))\n",
    "        if imgg is None:\n",
    "            continue\n",
    "        img = cv2.imread(path.join(base_path, file))\n",
    "#         cotton_ix += cotton_ix\n",
    "        print(len(cotton_ix))\n",
    "        tdata = [random.sample(bg_ix, 1 * len(cotton_ix)), cotton_ix]\n",
    "        colors = [(255, 0, 255), (0, 255, 255)]\n",
    "        for ixx in range(2):\n",
    "            for ix in tdata[ixx]:\n",
    "                val = training_data[ix]\n",
    "                samp_img = img[val[0]: val[1], val[2]: val[3]]\n",
    "                if do_resize:\n",
    "                    samp_img = np.resize(samp_img, (prediction_width, prediction_width, 3))\n",
    "                X.append(samp_img)\n",
    "                Y.append(ixx)\n",
    "                imgg = cv2.rectangle(imgg, (val[2], val[0]), (val[3], val[1]), colors[ixx], 2)\n",
    "        cv2.imwrite(path.join(base_path, \"training\", file), imgg)\n",
    "        #break\n",
    "    #break\n",
    "X = np.array(X)\n",
    "Y = np.array(Y)\n",
    "Y = to_categorical(Y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 225,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "/mnt/d/personal/cotton/Annoted_boll_burst_2019_trial/Set1\n",
      "2019_1206_122944_171.JPG\n",
      "399\n",
      "2019_1206_122945_172.JPG\n",
      "773\n",
      "2019_1206_122947_173.JPG\n",
      "733\n",
      "2019_1206_122948_174.JPG\n",
      "910\n",
      "2019_1206_122950_175.JPG\n",
      "1168\n",
      "2019_1206_122951_176.JPG\n",
      "998\n",
      "2019_1206_122953_177.JPG\n",
      "437\n",
      "2019_1206_122954_178.JPG\n",
      "496\n",
      "/mnt/d/personal/cotton/Annoted_boll_burst_2019_trial/Set2\n",
      "2019_1206_122959_181.JPG\n",
      "214\n",
      "2019_1206_123000_182.JPG\n",
      "170\n",
      "2019_1206_123002_183.JPG\n",
      "269\n",
      "2019_1206_123003_184.JPG\n",
      "409\n",
      "2019_1206_123005_185.JPG\n",
      "332\n",
      "2019_1206_123006_186.JPG\n",
      "290\n",
      "2019_1206_123008_187.JPG\n",
      "201\n",
      "2019_1206_123010_188.JPG\n",
      "189\n",
      "2019_1206_123011_189.JPG\n",
      "157\n",
      "2019_1206_123012_190.JPG\n",
      "286\n",
      "/mnt/d/personal/cotton/Annoted_boll_burst_2019_trial/Set3\n",
      "2019_1206_123014_191.JPG\n",
      "142\n",
      "2019_1206_123016_192.JPG\n",
      "202\n",
      "2019_1206_123017_193.JPG\n",
      "208\n",
      "2019_1206_123019_194.JPG\n",
      "150\n",
      "2019_1206_123020_195.JPG\n",
      "119\n",
      "2019_1206_123021_196.JPG\n",
      "267\n",
      "2019_1206_123023_197.JPG\n",
      "234\n",
      "2019_1206_123024_198.JPG\n",
      "206\n",
      "2019_1206_123026_199.JPG\n",
      "167\n",
      "2019_1206_123027_200.JPG\n",
      "145\n",
      "/mnt/d/personal/cotton/Annoted_boll_burst_2019_trial/Set4\n",
      "2019_1206_123029_201.JPG\n",
      "157\n",
      "2019_1206_123035_205.JPG\n",
      "164\n",
      "2019_1206_123036_206.JPG\n",
      "195\n",
      "2019_1206_123038_207.JPG\n",
      "545\n",
      "2019_1206_123039_208.JPG\n",
      "549\n",
      "2019_1206_123041_209.JPG\n",
      "693\n",
      "2019_1206_123042_210.JPG\n",
      "586\n",
      "2019_1206_123044_211.JPG\n",
      "511\n",
      "2019_1206_123045_212.JPG\n",
      "535\n",
      "2019_1206_123047_213.JPG\n",
      "475\n",
      "/mnt/d/personal/cotton/Annoted_boll_burst_2019_trial/Set5\n",
      "2019_1206_123048_214.JPG\n",
      "148\n",
      "2019_1206_123050_215.JPG\n",
      "674\n",
      "2019_1206_123051_216.JPG\n",
      "352\n",
      "2019_1206_123053_217.JPG\n",
      "282\n",
      "2019_1206_123054_218.JPG\n",
      "310\n",
      "2019_1206_123056_219.JPG\n",
      "133\n",
      "2019_1206_123057_220.JPG\n",
      "363\n",
      "2019_1206_123059_221.JPG\n",
      "290\n",
      "2019_1206_123100_222.JPG\n",
      "367\n",
      "2019_1206_123102_223.JPG\n",
      "778\n"
     ]
    }
   ],
   "source": [
    "# Create dataset using the annotated regions and resizing\n",
    "X = []\n",
    "Y = []\n",
    "folders = [\"Set1\", \"Set2\", \"Set3\", \"Set4\", \"Set5\"]\n",
    "test_images = [\"2019_1206_122956_179.JPG\", \"2019_1206_122957_180.JPG\"]\n",
    "stride = 20\n",
    "width = 40\n",
    "prediction_width = 40\n",
    "do_resize = True\n",
    "for folder in folders:\n",
    "    base_path = path.join(PATH, folder)\n",
    "    print(base_path)\n",
    "    ann = pd.read_csv(path.join(base_path, \"via_export_csv.csv\"))\n",
    "    for file in ann[\"filename\"].unique():\n",
    "        if file in test_images:\n",
    "            continue\n",
    "        cnt1 = 0\n",
    "        cnt0 = 0\n",
    "        print(file)\n",
    "        sample_ann = ann[ann[\"filename\"] == file]\n",
    "        img = cv2.imread(path.join(base_path, file))\n",
    "        if img is None:\n",
    "            continue\n",
    "        img1 = cv2.imread(path.join(base_path, file))\n",
    "        training_data = []\n",
    "\n",
    "        # Create a list of all the boxes\n",
    "        for _x in range(0, img.shape[1] - width, stride):\n",
    "            for _y in range(0, img.shape[0] - width, stride):\n",
    "                xstart = _x\n",
    "                xend = _x + width\n",
    "                ystart = _y\n",
    "                yend = _y + width\n",
    "                training_data.append([ystart, yend, xstart, xend])\n",
    "                \n",
    "        imgg = cv2.imread(path.join(base_path, \"labels\", file))\n",
    "        if imgg is None:\n",
    "            continue \n",
    "        colors = [(255, 0, 255), (0, 255, 255)]\n",
    "        # Loop through all annotations and fill the cotton pixels with 1\n",
    "        for ix, val in sample_ann.iterrows():\n",
    "            coords = json.loads(val[\"region_shape_attributes\"])\n",
    "            if \"x\" not in coords.keys():\n",
    "                continue\n",
    "            x1 = coords[\"x\"]\n",
    "            if x1 is None:\n",
    "                continue\n",
    "            x2 = x1 + coords[\"width\"]\n",
    "            y1 = coords[\"y\"]\n",
    "            y2 = y1 + coords[\"height\"]\n",
    "            img_crop = img[y1:y2, x1:x2]\n",
    "            if do_resize:\n",
    "                samp_img = cv2.resize(img_crop, (prediction_width, prediction_width), interpolation=cv2.INTER_AREA)\n",
    "            X.append(samp_img)\n",
    "            Y.append(1)\n",
    "            imgg = cv2.rectangle(imgg, (x1, y1), (x2, y2), colors[1], 2)\n",
    "            img[y1:y2, x1:x2] = np.ones_like(img_crop) * 200\n",
    " \n",
    "        # Find the ones with the fill less than a value\n",
    "        cotton_ix = []\n",
    "        bg_ix = []      \n",
    "        for ix in range(len(training_data)):\n",
    "            val = training_data[ix]\n",
    "            samp_img = img[val[0]: val[1], val[2]: val[3]]\n",
    "            if len(np.where(samp_img==200)[0]) > 4700:\n",
    "                cotton_ix.append(ix)\n",
    "            else:\n",
    "                bg_ix.append(ix)\n",
    "    \n",
    "\n",
    "        img = cv2.imread(path.join(base_path, file))\n",
    "#         cotton_ix += cotton_ix\n",
    "        print(len(cotton_ix))\n",
    "        tdata = [random.sample(bg_ix, 10 * len(cotton_ix)), cotton_ix]\n",
    "\n",
    "        for ixx in range(1):\n",
    "            for ix in tdata[ixx]:\n",
    "                val = training_data[ix]\n",
    "                samp_img = img[val[0]: val[1], val[2]: val[3]]\n",
    "                if do_resize:\n",
    "                    samp_img = cv2.resize(samp_img, (prediction_width, prediction_width))\n",
    "                X.append(samp_img)\n",
    "                Y.append(ixx)\n",
    "                imgg = cv2.rectangle(imgg, (val[2], val[0]), (val[3], val[1]), colors[ixx], 2)\n",
    "        cv2.imwrite(path.join(base_path, \"training\", file), imgg)\n",
    "        #break\n",
    "    # break\n",
    "X = np.array(X)\n",
    "Y = np.array(Y)\n",
    "Y = to_categorical(Y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 178,
   "metadata": {},
   "outputs": [],
   "source": [
    "def show(img):\n",
    "    plt.imshow(img[:, :, ::-1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 227,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "show(X[50])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "/mnt/d/personal/cotton/1\n",
      "2019_1010_123203_172.JPG\n",
      "2019_1010_123204_173.JPG\n",
      "2019_1010_123206_174.JPG\n",
      "2019_1010_123207_175.JPG\n",
      "2019_1010_123209_176.JPG\n",
      "2019_1010_123210_177.JPG\n",
      "2019_1010_123212_178.JPG\n",
      "2019_1010_123213_179.JPG\n",
      "2019_1010_123215_180.JPG\n",
      "2019_1010_123216_181.JPG\n",
      "2019_1010_123218_182.JPG\n",
      "2019_1010_123219_183.JPG\n",
      "2019_1010_123221_184.JPG\n",
      "2019_1010_123222_185.JPG\n",
      "2019_1010_123224_186.JPG\n",
      "2019_1010_123225_187.JPG\n",
      "2019_1010_123227_188.JPG\n",
      "2019_1010_123228_189.JPG\n",
      "2019_1010_123230_190.JPG\n",
      "2019_1010_123237_195.JPG\n",
      "2019_1010_123239_196.JPG\n",
      "2019_1010_123240_197.JPG\n",
      "2019_1010_123242_198.JPG\n",
      "2019_1010_123243_199.JPG\n",
      "2019_1010_123245_200.JPG\n",
      "2019_1010_123247_201.JPG\n",
      "2019_1010_123248_202.JPG\n",
      "2019_1010_123250_203.JPG\n",
      "2019_1010_123251_204.JPG\n",
      "2019_1010_123253_205.JPG\n",
      "2019_1010_123254_206.JPG\n",
      "2019_1010_123256_207.JPG\n",
      "2019_1010_123257_208.JPG\n",
      "2019_1010_123259_209.JPG\n",
      "2019_1010_123300_210.JPG\n",
      "2019_1010_123302_211.JPG\n",
      "2019_1010_123303_212.JPG\n",
      "2019_1010_123305_213.JPG\n",
      "2019_1010_123306_214.JPG\n",
      "2019_1010_123307_215.JPG\n",
      "2019_1010_123309_216.JPG\n",
      "2019_1010_123311_217.JPG\n",
      "2019_1010_123312_218.JPG\n",
      "2019_1010_123314_219.JPG\n",
      "2019_1010_123315_220.JPG\n",
      "2019_1010_123317_221.JPG\n",
      "2019_1010_123318_222.JPG\n",
      "2019_1010_123320_223.JPG\n",
      "2019_1010_123321_224.JPG\n",
      "2019_1010_123323_225.JPG\n",
      "/mnt/d/personal/cotton/2\n",
      "2019_1010_123327_228.JPG\n",
      "2019_1010_123328_229.JPG\n",
      "2019_1010_123330_230.JPG\n",
      "2019_1010_123332_231.JPG\n",
      "2019_1010_123333_232.JPG\n",
      "2019_1010_123335_233.JPG\n",
      "2019_1010_123336_234.JPG\n",
      "2019_1010_123338_235.JPG\n",
      "2019_1010_123339_236.JPG\n",
      "2019_1010_123341_237.JPG\n",
      "2019_1010_123342_238.JPG\n",
      "2019_1010_123344_239.JPG\n",
      "2019_1010_123345_240.JPG\n",
      "2019_1010_123347_241.JPG\n",
      "2019_1010_123348_242.JPG\n",
      "2019_1010_123353_245.JPG\n",
      "2019_1010_123354_246.JPG\n",
      "2019_1010_123356_247.JPG\n",
      "2019_1010_123357_248.JPG\n",
      "2019_1010_123359_249.JPG\n",
      "2019_1010_123400_250.JPG\n",
      "2019_1010_123402_251.JPG\n",
      "2019_1010_123403_252.JPG\n",
      "2019_1010_123405_253.JPG\n",
      "2019_1010_123406_254.JPG\n",
      "2019_1010_123407_255.JPG\n",
      "2019_1010_123409_256.JPG\n",
      "2019_1010_123410_257.JPG\n",
      "2019_1010_123416_261.JPG\n",
      "2019_1010_123418_262.JPG\n",
      "2019_1010_123420_263.JPG\n",
      "2019_1010_123421_264.JPG\n",
      "2019_1010_123423_265.JPG\n",
      "2019_1010_123424_266.JPG\n",
      "2019_1010_123426_267.JPG\n",
      "2019_1010_123427_268.JPG\n",
      "2019_1010_123429_269.JPG\n",
      "2019_1010_123430_270.JPG\n",
      "2019_1010_123432_271.JPG\n",
      "2019_1010_123433_272.JPG\n",
      "2019_1010_123435_273.JPG\n",
      "2019_1010_123436_274.JPG\n",
      "2019_1010_123438_275.JPG\n",
      "2019_1010_123439_276.JPG\n",
      "2019_1010_123441_277.JPG\n",
      "2019_1010_123442_278.JPG\n",
      "2019_1010_123444_279.JPG\n",
      "2019_1010_123445_280.JPG\n",
      "/mnt/d/personal/cotton/samples\n",
      "2019_1010_123336_234.JPG\n",
      "2019_1010_123338_235.JPG\n",
      "2019_1010_123339_236.JPG\n",
      "2019_1010_123341_237.JPG\n",
      "2019_1010_123342_238.JPG\n",
      "2019_1010_123344_239.JPG\n",
      "2019_1010_123345_240.JPG\n",
      "2019_1010_123347_241.JPG\n",
      "2019_1010_123348_242.JPG\n",
      "2019_1010_123353_245.JPG\n",
      "2019_1010_123354_246.JPG\n",
      "2019_1010_123356_247.JPG\n",
      "2019_1010_123357_248.JPG\n",
      "2019_1010_123359_249.JPG\n",
      "2019_1010_123400_250.JPG\n",
      "2019_1010_123402_251.JPG\n",
      "2019_1010_123403_252.JPG\n",
      "2019_1010_123405_253.JPG\n",
      "2019_1010_123406_254.JPG\n",
      "2019_1010_123407_255.JPG\n",
      "2019_1010_123409_256.JPG\n",
      "2019_1010_123410_257.JPG\n",
      "2019_1010_123416_261.JPG\n",
      "2019_1010_123418_262.JPG\n",
      "2019_1010_123420_263.JPG\n",
      "2019_1010_123421_264.JPG\n",
      "2019_1010_123423_265.JPG\n",
      "2019_1010_123424_266.JPG\n",
      "2019_1010_123426_267.JPG\n",
      "2019_1010_123427_268.JPG\n",
      "2019_1010_123429_269.JPG\n",
      "2019_1010_123430_270.JPG\n",
      "2019_1010_123432_271.JPG\n",
      "2019_1010_123433_272.JPG\n",
      "2019_1010_123435_273.JPG\n",
      "2019_1010_123436_274.JPG\n",
      "2019_1010_123438_275.JPG\n",
      "2019_1010_123439_276.JPG\n",
      "2019_1010_123441_277.JPG\n",
      "2019_1010_123442_278.JPG\n",
      "2019_1010_123444_279.JPG\n",
      "2019_1010_123445_280.JPG\n",
      "2019_1010_123508_295.JPG\n",
      "2019_1010_123509_296.JPG\n",
      "2019_1010_123511_297.JPG\n",
      "2019_1010_123512_298.JPG\n",
      "2019_1010_123514_299.JPG\n",
      "2019_1010_123515_300.JPG\n",
      "2019_1010_123520_303.JPG\n",
      "2019_1010_123521_304.JPG\n",
      "2019_1010_123523_305.JPG\n",
      "2019_1010_123524_306.JPG\n",
      "2019_1010_123525_307.JPG\n",
      "2019_1010_123527_308.JPG\n",
      "2019_1010_123528_309.JPG\n",
      "2019_1010_123530_310.JPG\n",
      "2019_1010_123532_311.JPG\n",
      "2019_1010_123533_312.JPG\n",
      "2019_1010_123535_313.JPG\n",
      "2019_1010_123536_314.JPG\n",
      "2019_1010_123538_315.JPG\n",
      "2019_1010_123539_316.JPG\n",
      "[1]2019_1010_123336_234.JPG\n"
     ]
    }
   ],
   "source": [
    "# Create a dataset using mis-predictions\n",
    "Predict = True\n",
    "X = []\n",
    "Y = []\n",
    "folders = [\"1\", \"2\", \"samples\"]\n",
    "# folders = [\"samples\"]\n",
    "test_images = [\"2019_1010_123350_243.JPG\", \"2019_1010_123351_244.JPG\"]\n",
    "for folder in folders:\n",
    "    base_path = path.join(PATH, folder)\n",
    "    print(base_path)\n",
    "    ann = pd.read_csv(path.join(base_path, \"via_export_csv.csv\"))\n",
    "    for file in ann[\"filename\"].unique():\n",
    "        if file in test_images:\n",
    "            continue\n",
    "#         if file != \"2019_1010_123336_234.JPG\":\n",
    "#             continue\n",
    "#         if file != \"2019_1010_123336_234_7_4.jpg\":\n",
    "#             continue\n",
    "        training_data = []\n",
    "        print(file)\n",
    "        \n",
    "        # Create a list of all 40 40 boxes with 20 stride\n",
    "        sample_ann = ann[ann[\"filename\"] == file]\n",
    "        img = cv2.imread(path.join(base_path, file))\n",
    "        if img is None:\n",
    "            continue\n",
    "        for _x in range(0, img.shape[1] - 40, 20):\n",
    "            for _y in range(0, img.shape[0] - 40, 20):\n",
    "                xstart = _x\n",
    "                xend = _x + 40\n",
    "                ystart = _y\n",
    "                yend = _y + 40\n",
    "                training_data.append([ystart, yend, xstart, xend])\n",
    "        \n",
    "        cnt = 0\n",
    "        # Read all the annotations\n",
    "        for ix, val in sample_ann.iterrows():\n",
    "            coords = json.loads(val[\"region_shape_attributes\"])\n",
    "            if \"x\" not in coords.keys():\n",
    "                continue\n",
    "            x1 = coords[\"x\"]\n",
    "            if x1 is None:\n",
    "                continue\n",
    "            x2 = x1 + coords[\"width\"]\n",
    "            y1 = coords[\"y\"]\n",
    "            y2 = y1 + coords[\"height\"]\n",
    "            img_crop = img[y1: y2, x1:x2]\n",
    "            X.append(cv2.resize(img_crop, (40, 40)))\n",
    "            Y.append(1)\n",
    "            X.append(cv2.resize(img_crop, (40, 40)))\n",
    "            Y.append(1)\n",
    "            cnt += 1\n",
    "            \n",
    "        if cnt == 0:\n",
    "            continue\n",
    "        \n",
    "  \n",
    "        # Predict all the boxes using the current model\n",
    "        if Predict:\n",
    "            all_images = []\n",
    "            predictions = []\n",
    "#             for ix in tqdm.tqdm_notebook(range(len(training_data))):\n",
    "            for ix in range(len(training_data)):\n",
    "                val = training_data[ix]\n",
    "                samp_img = img[val[0]: val[1], val[2]: val[3]]\n",
    "                all_images.append(samp_img)\n",
    "                if len(all_images) == 50000:\n",
    "                    all_images = np.array(all_images) * 1.0 / 255\n",
    "                    temp_predictions = model.predict(all_images, verbose=1).tolist()\n",
    "                    predictions.extend(temp_predictions)\n",
    "                    all_images = []\n",
    "            all_images = np.array(all_images) * 1.0 / 255\n",
    "            predictions.extend(model.predict(all_images, verbose =0).tolist())\n",
    "        all_boxes = []\n",
    "        all_scores = []\n",
    "        for ix in range(len(predictions)):\n",
    "            prob = predictions[ix][1]\n",
    "            if prob > 0.5:\n",
    "                val = training_data[ix]\n",
    "                all_boxes.append([val[2], val[0], 40, 40])\n",
    "                all_scores.append(prob)\n",
    "        indices = nms.boxes(all_boxes, all_scores, nms_threshold=0.1)\n",
    "\n",
    "        for ix, val in sample_ann.iterrows():\n",
    "            coords = json.loads(val[\"region_shape_attributes\"])\n",
    "            if \"x\" not in coords.keys():\n",
    "                continue\n",
    "            x1 = coords[\"x\"]\n",
    "            if x1 is None:\n",
    "                continue\n",
    "            x2 = x1 + coords[\"width\"]\n",
    "            y1 = coords[\"y\"]\n",
    "            y2 = y1 + coords[\"height\"]\n",
    "            img_crop = img[y1:y2, x1:x2]\n",
    "            img[y1:y2, x1:x2] = np.ones_like(img_crop)\n",
    "        imgg = cv2.imread(path.join(base_path, \"labels\", file))   \n",
    "        for ix in indices:\n",
    "            if all_scores[ix] > 0.9:\n",
    "                x1 = all_boxes[ix][0]\n",
    "                y1 = all_boxes[ix][1]\n",
    "                x2 = x1 + 40\n",
    "                y2 = y1 + 40\n",
    "                img_crop = img[y1:y2, x1:x2]\n",
    "#                 print(x1, x2, y1, y2)\n",
    "                if len(np.where(img_crop==1)[0]) < 100:\n",
    "#                 if True:\n",
    "                    X.append(img_crop.copy())\n",
    "                    Y.append(0)\n",
    "                    imgg = cv2.rectangle(imgg, (x1, y1), (x2, y2), (0, 255, 0), 2)\n",
    "                else:\n",
    "                    imgg = cv2.rectangle(imgg, (x1, y1), (x2, y2), (0, 255, 255), 2)\n",
    "        if imgg is not None:\n",
    "            cv2.imwrite(path.join(base_path, \"predictions\", file), imgg)\n",
    "            \n",
    "X = np.array(X)\n",
    "Y = np.array(Y)\n",
    "Y = to_categorical(Y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(28149, 2)"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Y.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "2559"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.count_nonzero(np.argmax(Y, axis=1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 536,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "139\n"
     ]
    }
   ],
   "source": [
    "# Runs the model through the sample images and draws a bounding box\n",
    "do_nms = True\n",
    "do_prediction = False\n",
    "prob_thresh = 0.5\n",
    "stride = 20\n",
    "base_path = path.join(PATH, \"samples\")\n",
    "cnt = 0\n",
    "ann = pd.read_csv(path.join(base_path, \"via_export_csv.csv\"))\n",
    "for file in ann[\"filename\"].unique():\n",
    "    if file != \"2019_1010_123350_243.JPG\":\n",
    "        continue\n",
    "#     if not file.startswith(\"2019_1010_123350_243\"):\n",
    "#         continue\n",
    "    sample_ann = ann[ann[\"filename\"] == file]\n",
    "    training_data = []\n",
    "    img = cv2.imread(path.join(base_path, file))\n",
    "    for _x in range(0, img.shape[1] - 40, stride):\n",
    "        for _y in range(0, img.shape[0] - 40, stride):\n",
    "            xstart = _x\n",
    "            xend = _x + 40\n",
    "            ystart = _y\n",
    "            yend = _y + 40\n",
    "            training_data.append([ystart, yend, xstart, xend])\n",
    "    \n",
    "    img = cv2.imread(path.join(base_path, file))\n",
    "    if do_prediction:\n",
    "        all_images = []\n",
    "        predictions = []\n",
    "        for ix in tqdm.tqdm_notebook(range(len(training_data))):\n",
    "            val = training_data[ix]\n",
    "            samp_img = img[val[0]: val[1], val[2]: val[3]]\n",
    "            all_images.append(samp_img)\n",
    "            if len(all_images) == 50000:\n",
    "                all_images = np.array(all_images) * 1.0 / 255\n",
    "                temp_predictions = model.predict_proba(all_images, verbose=1).tolist()\n",
    "                predictions.extend(temp_predictions)\n",
    "                all_images = []\n",
    "        all_images = np.array(all_images) * 1.0 / 255\n",
    "        predictions.extend(model.predict_proba(all_images, verbose =1).tolist())\n",
    "\n",
    "    img = cv2.imread(path.join(base_path, \"labels\", file))\n",
    "    all_boxes = []\n",
    "    all_scores = []\n",
    "    for ix in range(len(predictions)):\n",
    "        prob = predictions[ix][1]\n",
    "        if prob > 0.5:\n",
    "            val = training_data[ix]\n",
    "            all_boxes.append([val[2], val[0], 40, 40])\n",
    "            all_scores.append(prob)\n",
    "    if do_nms:\n",
    "        indices = nms.boxes(all_boxes, all_scores, nms_threshold=0.1)\n",
    "    else:\n",
    "        indices = range(len(all_boxes))\n",
    "     \n",
    "    for ix in indices:\n",
    "        if all_scores[ix] > prob_thresh:\n",
    "            cnt += 1\n",
    "            img = cv2.rectangle(img, (all_boxes[ix][0], all_boxes[ix][1]), (all_boxes[ix][0] + 40, all_boxes[ix][1] + 40), (0, 255, 0), 2)\n",
    "    cv2.imwrite(path.join(base_path, \"predictions\", file), img)\n",
    "    break\n",
    "print(cnt)"
   ]
  }
 ],
 "metadata": {
  "celltoolbar": "Raw Cell Format",
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}