Basset/train_eugene.ipynb

{
 "cells": [
  {
   "attachments": {},
   "cell_type": "markdown",
   "id": "8ec3509e",
   "metadata": {},
   "source": [
    "# Retraining Basset with EUGENe\n",
    "**Authorship:**\n",
    "Adam Klie (last updated: *07/19/2023*)\n",
    "***\n",
    "**Description:**\n",
    "Test the Kipoi hosted implementation of Basenji for inference on some randomly generated sequences. This is a good place to check if your installation is working properly.\n",
    "***"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "id": "c91addbb-00f9-474a-8386-b61bf03af5bd",
   "metadata": {
    "id": "SVxwvzYj-txy"
   },
   "source": [
    "# Set-up"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "5256fba9",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2023-03-27T20:06:52.328546Z",
     "iopub.status.busy": "2023-03-27T20:06:52.328175Z",
     "iopub.status.idle": "2023-03-27T20:07:20.113060Z",
     "shell.execute_reply": "2023-03-27T20:07:20.112385Z",
     "shell.execute_reply.started": "2023-03-27T20:06:52.328498Z"
    }
   },
   "outputs": [],
   "source": [
    "# General imports\n",
    "import os\n",
    "import time\n",
    "import h5py\n",
    "import numpy as np"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "id": "a9a29fa3-f880-4dfb-8b41-2eec95e9f5ad",
   "metadata": {
    "id": "VuUNVfty-tx4"
   },
   "source": [
    "# Load dataset from EvoAug publication"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "59808aeb-e3be-4835-a34d-3ef8e5acf704",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2023-03-27T20:07:20.122078Z",
     "iopub.status.busy": "2023-03-27T20:07:20.121720Z",
     "iopub.status.idle": "2023-03-27T20:07:20.138122Z",
     "shell.execute_reply": "2023-03-27T20:07:20.137504Z",
     "shell.execute_reply.started": "2023-03-27T20:07:20.122060Z"
    },
    "id": "wl8UGRA6-tyA",
    "tags": []
   },
   "outputs": [],
   "source": [
    "dataset_path = \"/cellar/users/aklie/projects/ML4GLand/data/pubs/lee22/Basset_data.h5\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "f2362b84",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load dataset from h5 file\n",
    "with h5py.File(dataset_path, 'r') as dataset:\n",
    "    X_train = np.array(dataset['X_train'])\n",
    "    y_train = np.array(dataset['Y_train'])\n",
    "    X_valid = np.array(dataset['X_valid'])\n",
    "    y_valid = np.array(dataset['Y_valid'])\n",
    "    X_test = np.array(dataset['X_test'])\n",
    "    y_test = np.array(dataset['Y_test'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "a23cf1a6",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((437478, 4, 600),\n",
       " (437478, 164),\n",
       " (16410, 4, 600),\n",
       " (16410, 164),\n",
       " (16703, 4, 600),\n",
       " (16703, 164))"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Check the shape of the data\n",
    "X_train.shape, y_train.shape, X_valid.shape, y_valid.shape, X_test.shape, y_test.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "366be40c",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(4.1997888, 4.199788944)"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Check training dataset size in mem (GBs)\n",
    "import sys\n",
    "X_train.nbytes / 1e9, sys.getsizeof(X_train) / 1e9"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "id": "80d8abef-62d6-47be-90d9-b02ddb55fec9",
   "metadata": {
    "id": "JsllXGpZ-tyB"
   },
   "source": [
    "# Instantiate model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "4ab1f512-1f1b-49d7-9c1a-5d74bb63b29d",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T01:27:50.590512Z",
     "iopub.status.busy": "2022-12-20T01:27:50.590297Z",
     "iopub.status.idle": "2022-12-20T01:27:53.863495Z",
     "shell.execute_reply": "2022-12-20T01:27:53.863001Z",
     "shell.execute_reply.started": "2022-12-20T01:27:50.590495Z"
    },
    "id": "fwxajn6z-tyB",
    "tags": []
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "OMP: Info #276: omp_set_nested routine deprecated, please use omp_set_max_active_levels instead.\n"
     ]
    }
   ],
   "source": [
    "from eugene.models.zoo import Basset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "9d4e6b42-c0c0-418f-abba-a0e89c609dbf",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T01:59:35.591195Z",
     "iopub.status.busy": "2022-12-20T01:59:35.590850Z",
     "iopub.status.idle": "2022-12-20T01:59:40.255932Z",
     "shell.execute_reply": "2022-12-20T01:59:40.255407Z",
     "shell.execute_reply.started": "2022-12-20T01:59:35.591176Z"
    },
    "id": "lXALE6P6-tyB",
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Basset(\n",
       "  (conv1d_tower): Conv1DTower(\n",
       "    (layers): Sequential(\n",
       "      (0): Conv1d(4, 300, kernel_size=(19,), stride=(1,), padding=(9,))\n",
       "      (1): BatchNorm1d(300, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (2): ReLU()\n",
       "      (3): MaxPool1d(kernel_size=3, stride=1, padding=0, dilation=1, ceil_mode=False)\n",
       "      (4): Conv1d(300, 200, kernel_size=(11,), stride=(1,), padding=(5,))\n",
       "      (5): BatchNorm1d(200, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (6): ReLU()\n",
       "      (7): MaxPool1d(kernel_size=4, stride=1, padding=0, dilation=1, ceil_mode=False)\n",
       "      (8): Conv1d(200, 200, kernel_size=(7,), stride=(1,), padding=(3,))\n",
       "      (9): BatchNorm1d(200, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (10): ReLU()\n",
       "      (11): MaxPool1d(kernel_size=4, stride=1, padding=0, dilation=1, ceil_mode=False)\n",
       "    )\n",
       "  )\n",
       "  (dense_block): DenseBlock(\n",
       "    (layers): Sequential(\n",
       "      (0): Linear(in_features=118400, out_features=1000, bias=True)\n",
       "      (1): BatchNorm1d(1000, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (2): ReLU()\n",
       "      (3): Linear(in_features=1000, out_features=164, bias=True)\n",
       "      (4): BatchNorm1d(164, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (5): ReLU()\n",
       "      (6): Linear(in_features=164, out_features=164, bias=True)\n",
       "    )\n",
       "  )\n",
       ")"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "arch = Basset(\n",
    "    input_len=600, \n",
    "    output_dim=164\n",
    ")\n",
    "arch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "ec6b9259",
   "metadata": {},
   "outputs": [],
   "source": [
    "from eugene.models import SequenceModule"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "452810e8",
   "metadata": {},
   "outputs": [],
   "source": [
    "model = SequenceModule(\n",
    "    arch=arch,\n",
    "    task=\"multilabel_classification\",\n",
    "    loss_fxn=\"bce\",\n",
    "    optimizer=\"adam\",\n",
    "    metric_kwargs={\n",
    "        \"task\": \"multilabel\",\n",
    "        \"num_labels\": 164\n",
    "    }\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "d3ce8d36",
   "metadata": {},
   "outputs": [],
   "source": [
    "from eugene.models import init_weights"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "8fccc503",
   "metadata": {},
   "outputs": [],
   "source": [
    "init_weights(model)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "1c361738-65d2-400c-aa5d-c9881143366a",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T02:00:11.084802Z",
     "iopub.status.busy": "2022-12-20T02:00:11.084156Z",
     "iopub.status.idle": "2022-12-20T02:00:14.432587Z",
     "shell.execute_reply": "2022-12-20T02:00:14.431976Z",
     "shell.execute_reply.started": "2022-12-20T02:00:11.084782Z"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: Basset\n",
      "Sequence length: 600\n",
      "Output dimension: 164\n",
      "Task: multilabel_classification\n",
      "Loss function: binary_cross_entropy_with_logits\n",
      "Optimizer: Adam\n",
      "\tOptimizer parameters: {}\n",
      "\tOptimizer starting learning rate: 0.001\n",
      "Scheduler: None\n",
      "\tScheduler parameters: {}\n",
      "Metric: auroc\n",
      "\tMetric parameters: {'task': 'multilabel', 'num_labels': 164}\n",
      "Seed: None\n",
      "Parameters summary:\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "  | Name         | Type            | Params\n",
       "-------------------------------------------------\n",
       "0 | arch         | Basset          | 119 M \n",
       "1 | train_metric | MultilabelAUROC | 0     \n",
       "2 | val_metric   | MultilabelAUROC | 0     \n",
       "3 | test_metric  | MultilabelAUROC | 0     \n",
       "-------------------------------------------------\n",
       "119 M     Trainable params\n",
       "0         Non-trainable params\n",
       "119 M     Total params\n",
       "478.238   Total estimated model params size (MB)"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "id": "0d368bf2-4cbe-47c7-9752-1c59ad270251",
   "metadata": {
    "id": "1hAUZjIs-tyC"
   },
   "source": [
    "# Build dataloader"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "add6d470",
   "metadata": {},
   "outputs": [],
   "source": [
    "import seqdata as sd\n",
    "import xarray as xr"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "4001c01e",
   "metadata": {},
   "outputs": [],
   "source": [
    "train_sdata = xr.Dataset(\n",
    "    {\n",
    "        \"ohe_seq\": ([\"_sequence\", \"_ohe\", \"_length\"], X_train),\n",
    "        \"target\": ([\"_sequence\", \"_targets\"], y_train),\n",
    "    }\n",
    ")\n",
    "\n",
    "valid_sdata = xr.Dataset(\n",
    "    {\n",
    "        \"ohe_seq\": ([\"_sequence\", \"_ohe\", \"_length\"], X_valid),\n",
    "        \"target\": ([\"_sequence\", \"_targets\"], y_valid),\n",
    "\n",
    "    }\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "aa4b454c",
   "metadata": {},
   "outputs": [],
   "source": [
    "train_dl = sd.get_torch_dataloader(\n",
    "    train_sdata,\n",
    "    batch_size=100,\n",
    "    sample_dims=\"_sequence\",\n",
    "    variables=[\"ohe_seq\", \"target\"],\n",
    "    shuffle=True,\n",
    "    num_workers=4,\n",
    "    prefetch_factor=2,\n",
    ")\n",
    "\n",
    "val_dl = sd.get_torch_dataloader(\n",
    "    valid_sdata,\n",
    "    batch_size=100,\n",
    "    sample_dims=\"_sequence\",\n",
    "    variables=[\"ohe_seq\", \"target\"],\n",
    "    shuffle=False,\n",
    "    num_workers=4,\n",
    "    prefetch_factor=2,\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "bb4c2753",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(torch.Size([100, 4, 600]), torch.Size([100, 164]))"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "batch = next(iter(train_dl))\n",
    "batch[\"ohe_seq\"].shape, batch[\"target\"].shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "fb813d7e",
   "metadata": {},
   "outputs": [],
   "source": [
    "from eugene import train"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "id": "43730729",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[rank: 0] Global seed set to 13\n",
      "/cellar/users/aklie/opt/miniconda3/envs/ml4gland/lib/python3.9/site-packages/lightning_fabric/plugins/environments/slurm.py:165: PossibleUserWarning: The `srun` command is available on your system but is not used. HINT: If your intention is to run Lightning on SLURM, prepend your python command with `srun` like so: srun python /cellar/users/aklie/opt/miniconda3/envs/ml4gland/lib ...\n",
      "  rank_zero_warn(\n",
      "GPU available: True (cuda), used: True\n",
      "TPU available: False, using: 0 TPU cores\n",
      "IPU available: False, using: 0 IPUs\n",
      "HPU available: False, using: 0 HPUs\n",
      "/cellar/users/aklie/opt/miniconda3/envs/ml4gland/lib/python3.9/site-packages/lightning_fabric/plugins/environments/slurm.py:165: PossibleUserWarning: The `srun` command is available on your system but is not used. HINT: If your intention is to run Lightning on SLURM, prepend your python command with `srun` like so: srun python /cellar/users/aklie/opt/miniconda3/envs/ml4gland/lib ...\n",
      "  rank_zero_warn(\n",
      "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n",
      "\n",
      "  | Name         | Type            | Params\n",
      "-------------------------------------------------\n",
      "0 | arch         | Basset          | 119 M \n",
      "1 | train_metric | MultilabelAUROC | 0     \n",
      "2 | val_metric   | MultilabelAUROC | 0     \n",
      "3 | test_metric  | MultilabelAUROC | 0     \n",
      "-------------------------------------------------\n",
      "119 M     Trainable params\n",
      "0         Non-trainable params\n",
      "119 M     Total params\n",
      "478.238   Total estimated model params size (MB)\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "0e1900b1383c43f580730e8ef0428a86",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Sanity Checking: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "e650f44c7f8d49949715b764c6534eb2",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Training: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/cellar/users/aklie/opt/miniconda3/envs/ml4gland/lib/python3.9/site-packages/torchmetrics/utilities/prints.py:36: UserWarning: No positive samples in targets, true positive value should be meaningless. Returning zero tensor in true positive score\n",
      "  warnings.warn(*args, **kwargs)\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "7e3d9453f62f413a9df7e385ff4d3de0",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validation: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "a79a1da6e25a477c86bb92f75fdcd2df",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validation: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "6c38f1190a7146ed9920240458f9cd90",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validation: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "0ad646f38acb41ffa9453e388e094bef",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validation: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "10a53373d7f4467cab8cf805002ffe9a",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validation: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "8c67b33537ce4ef7961a94e2d60c4076",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validation: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "9980a1e5f632419f968de40f6430abc8",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validation: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "a2d8a00c812946c78d957d9edfd0b7d3",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validation: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "bfcd07ecf5a64ec699a165b0a37085ab",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validation: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "39dddfb9dcb649838859a3b058aaf698",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validation: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "4847f72ddea040579959db298c7de639",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validation: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "eef8f219970a4331a23d72fd58029e2a",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validation: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "5b89b4ff897446bf913900acc120c54d",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Validation: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "train.fit(\n",
    "    model=model,\n",
    "    train_dataloader=train_dl,\n",
    "    val_dataloader=val_dl,\n",
    "    epochs=100,\n",
    "    gpus=1,\n",
    "    logger=\"tensorboard\",\n",
    "    log_dir=\"/cellar/users/aklie/projects/ML4GLand/models/\",  # TODO: change to your own path\n",
    "    name=\"Basset\",\n",
    "    version=time.strftime(\"%Y-%m-%d_%H-%M-%S\"),\n",
    "    early_stopping_patience=10,\n",
    "    seed=13\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "37055696-d53f-47b3-bfe1-e499ada5b168",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T03:59:28.037764Z",
     "iopub.status.busy": "2022-12-20T03:59:28.037576Z",
     "iopub.status.idle": "2022-12-20T03:59:31.537648Z",
     "shell.execute_reply": "2022-12-20T03:59:31.536931Z",
     "shell.execute_reply.started": "2022-12-20T03:59:28.037747Z"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "import pytorch_lightning as pl"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "id": "87850c72-dcfc-4b39-92ec-b2df58f796f4",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T01:46:50.116800Z",
     "iopub.status.busy": "2022-12-20T01:46:50.116506Z",
     "iopub.status.idle": "2022-12-20T01:46:52.305417Z",
     "shell.execute_reply": "2022-12-20T01:46:52.304867Z",
     "shell.execute_reply.started": "2022-12-20T01:46:50.116780Z"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "callback_topmodel = pl.callbacks.ModelCheckpoint(monitor='val_loss', \n",
    "                                                 save_top_k=1, \n",
    "                                                 dirpath=\"eugene_logs\", \n",
    "                                                 filename=\"Basset\")\n",
    "callback_es = pl.callbacks.early_stopping.EarlyStopping(monitor='val_loss', patience=10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "id": "4c764899-ab12-431e-b17e-e9889dd6b99c",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T01:46:52.306573Z",
     "iopub.status.busy": "2022-12-20T01:46:52.306269Z",
     "iopub.status.idle": "2022-12-20T01:46:54.748236Z",
     "shell.execute_reply": "2022-12-20T01:46:54.747817Z",
     "shell.execute_reply.started": "2022-12-20T01:46:52.306556Z"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "GPU available: True, used: True\n",
      "TPU available: False, using: 0 TPU cores\n",
      "IPU available: False, using: 0 IPUs\n"
     ]
    }
   ],
   "source": [
    "logger = pl.loggers.TensorBoardLogger(save_dir=\"eugene_logs\")\n",
    "trainer = pl.Trainer(gpus=1, max_epochs=100, auto_select_gpus=True, logger=logger, callbacks=[callback_es, callback_topmodel])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "2b83fb00-b7ed-4e9b-88c3-76780b465e6f",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T03:55:13.798098Z",
     "iopub.status.busy": "2022-12-20T03:55:13.797859Z",
     "iopub.status.idle": "2022-12-20T03:55:17.054469Z",
     "shell.execute_reply": "2022-12-20T03:55:17.053794Z",
     "shell.execute_reply.started": "2022-12-20T03:55:13.798080Z"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "# Fit to data\n",
    "trainer.fit(model, datamodule=data_module)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "157a26dc-7631-4a15-a9f4-ae5ea97cd4f5",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T03:57:23.624869Z",
     "iopub.status.busy": "2022-12-20T03:57:23.624321Z",
     "iopub.status.idle": "2022-12-20T03:57:28.995372Z",
     "shell.execute_reply": "2022-12-20T03:57:28.994720Z",
     "shell.execute_reply.started": "2022-12-20T03:57:23.624847Z"
    },
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 864x432 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "eu.pl.training_summary(\"eugene_logs/default/version_1/\", metric=\"auroc\")"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "id": "5ed66e40-ff07-445e-a34f-00c95177f764",
   "metadata": {},
   "source": [
    "# Evaluate model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "27804ef5-bf48-4c68-b5ad-faa4f492bb4e",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T03:56:25.031642Z",
     "iopub.status.busy": "2022-12-20T03:56:25.030948Z",
     "iopub.status.idle": "2022-12-20T03:56:28.636330Z",
     "shell.execute_reply": "2022-12-20T03:56:28.635833Z",
     "shell.execute_reply.started": "2022-12-20T03:56:25.031619Z"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "MODEL_PATH = \"/cellar/users/aklie/projects/EUGENe_use_cases/kelley16/eugene_logs/Basset.ckpt\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "2cdffc1a-3e45-460d-a878-b59cd62e93bc",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T03:56:28.639255Z",
     "iopub.status.busy": "2022-12-20T03:56:28.638789Z",
     "iopub.status.idle": "2022-12-20T03:56:35.120192Z",
     "shell.execute_reply": "2022-12-20T03:56:35.119578Z",
     "shell.execute_reply.started": "2022-12-20T03:56:28.639233Z"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "None {'num_classes': 164} multilabel_classification\n",
      "auroc\n",
      "{'num_classes': 164}\n",
      "AUROC()\n",
      "{'regression': {}, 'binary_classification': {'task': 'binary'}, 'multiclass_classification': {'task': 'multiclass'}, 'multilabel_classification': {'task': 'multilabel'}}\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/cellar/users/aklie/opt/miniconda3/envs/eugene_dev/lib/python3.7/site-packages/torchmetrics/utilities/prints.py:36: UserWarning: Metric `AUROC` will save all targets and predictions in buffer. For large datasets this may lead to large memory footprint.\n",
      "  warnings.warn(*args, **kwargs)\n"
     ]
    }
   ],
   "source": [
    "best_model = eu.models.Basset.load_from_checkpoint(MODEL_PATH)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "id": "b14043aa-d393-4ecc-9808-8eb18f674582",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T04:06:20.591684Z",
     "iopub.status.busy": "2022-12-20T04:06:20.591454Z",
     "iopub.status.idle": "2022-12-20T04:06:24.383120Z",
     "shell.execute_reply": "2022-12-20T04:06:24.382519Z",
     "shell.execute_reply.started": "2022-12-20T04:06:20.591666Z"
    },
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0., 0., 0., ..., 0., 0., 0.],\n",
       "       [0., 0., 0., ..., 0., 0., 0.],\n",
       "       [0., 0., 0., ..., 0., 0., 0.],\n",
       "       ...,\n",
       "       [0., 0., 0., ..., 0., 0., 0.],\n",
       "       [0., 0., 0., ..., 0., 0., 0.],\n",
       "       [0., 1., 0., ..., 0., 0., 0.]], dtype=float32)"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data_module.y_test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "86857f29-1786-4122-bb17-00a97752a34c",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T04:01:51.820802Z",
     "iopub.status.busy": "2022-12-20T04:01:51.820462Z",
     "iopub.status.idle": "2022-12-20T04:01:55.758017Z",
     "shell.execute_reply": "2022-12-20T04:01:55.757357Z",
     "shell.execute_reply.started": "2022-12-20T04:01:51.820772Z"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "from torchvision import transforms\n",
    "from eugene.dataload import ToTensor"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "id": "f4a4faee-91fe-40a1-9f1d-9e3bd596cd28",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T04:09:06.772052Z",
     "iopub.status.busy": "2022-12-20T04:09:06.771639Z",
     "iopub.status.idle": "2022-12-20T04:09:09.792858Z",
     "shell.execute_reply": "2022-12-20T04:09:09.792435Z",
     "shell.execute_reply.started": "2022-12-20T04:09:06.772031Z"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "sdataset = eu.dl.SeqDataset(\n",
    "    seqs=data_module.x_test, \n",
    "    targets=data_module.y_test, \n",
    "    names=np.array([f\"seq{i}\" for i in range(len(data_module.x_test))]),\n",
    "    transform=transforms.Compose([ToTensor()])\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "id": "ad29563d-e12b-4910-9ac5-287f923b1596",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T04:09:12.127892Z",
     "iopub.status.busy": "2022-12-20T04:09:12.127725Z",
     "iopub.status.idle": "2022-12-20T04:09:15.098386Z",
     "shell.execute_reply": "2022-12-20T04:09:15.097919Z",
     "shell.execute_reply.started": "2022-12-20T04:09:12.127875Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(tensor([115., 101., 113.,  48.,  36.,  36.,  36.,  36.]),\n",
       " tensor([[1., 0., 0.,  ..., 1., 0., 1.],\n",
       "         [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "         [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "         [0., 1., 1.,  ..., 0., 1., 0.]]),\n",
       " tensor([-1.]),\n",
       " tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "         0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 1., 0., 0., 0., 0.,\n",
       "         0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "         0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "         0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "         0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "         0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "         0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "         0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "         0., 0.]))"
      ]
     },
     "execution_count": 47,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sdataset[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "id": "43c8eee1-fb83-4789-ab65-a2a8a2040793",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T04:09:18.591849Z",
     "iopub.status.busy": "2022-12-20T04:09:18.591555Z",
     "iopub.status.idle": "2022-12-20T04:09:22.503988Z",
     "shell.execute_reply": "2022-12-20T04:09:22.503469Z",
     "shell.execute_reply.started": "2022-12-20T04:09:18.591829Z"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "sdataloader = sdataset.to_dataloader(batch_size=128, shuffle=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "id": "b65c4708-ca99-44a1-9c8e-0fbc301a233a",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T04:09:22.984679Z",
     "iopub.status.busy": "2022-12-20T04:09:22.984514Z",
     "iopub.status.idle": "2022-12-20T04:09:26.849138Z",
     "shell.execute_reply": "2022-12-20T04:09:26.848722Z",
     "shell.execute_reply.started": "2022-12-20T04:09:22.984663Z"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "GPU available: True, used: True\n",
      "TPU available: False, using: 0 TPU cores\n",
      "IPU available: False, using: 0 IPUs\n"
     ]
    }
   ],
   "source": [
    "trainer = pl.Trainer(gpus=1, auto_select_gpus=True, logger=None)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "id": "6e5efb10-7a38-4f8e-9f02-fb2d2810efe7",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T04:10:21.642952Z",
     "iopub.status.busy": "2022-12-20T04:10:21.642703Z",
     "iopub.status.idle": "2022-12-20T04:10:32.026569Z",
     "shell.execute_reply": "2022-12-20T04:10:32.026034Z",
     "shell.execute_reply.started": "2022-12-20T04:10:21.642934Z"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "3613753018894767abdcfbc22e3ca6e2",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Predicting: 0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "pred = np.concatenate(trainer.predict(best_model, dataloaders=sdataloader))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "id": "88de3931-4b03-4243-b74c-44e4b7db20fe",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T04:14:47.591159Z",
     "iopub.status.busy": "2022-12-20T04:14:47.590961Z",
     "iopub.status.idle": "2022-12-20T04:14:53.616885Z",
     "shell.execute_reply": "2022-12-20T04:14:53.616345Z",
     "shell.execute_reply.started": "2022-12-20T04:14:47.591142Z"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "preds = pred[:, 1:165].astype(np.float32)\n",
    "trues = pred[:, 165:].astype(np.float32)\n",
    "ids = pred[:, 1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "id": "4e7f7b6f-33d8-4ceb-aa8b-ad2de12b0e4f",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T04:14:55.028125Z",
     "iopub.status.busy": "2022-12-20T04:14:55.027814Z",
     "iopub.status.idle": "2022-12-20T04:14:58.812207Z",
     "shell.execute_reply": "2022-12-20T04:14:58.811739Z",
     "shell.execute_reply.started": "2022-12-20T04:14:55.028107Z"
    },
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[-4.5189333, -4.5125437, -3.7900558, ..., -3.0956502, -3.6418533,\n",
       "        -4.6703386],\n",
       "       [-2.5667593, -5.189077 , -2.7265296, ..., -3.2501156, -3.4317393,\n",
       "        -4.0454507],\n",
       "       [-6.031025 , -4.0759897, -5.158294 , ..., -3.22485  , -4.0770197,\n",
       "        -4.8552556],\n",
       "       ...,\n",
       "       [-2.4186454, -3.8015072, -2.6640704, ..., -2.250576 , -2.9372458,\n",
       "        -3.97238  ],\n",
       "       [-5.2373266, -4.0637956, -4.5207825, ..., -3.826011 , -4.3333616,\n",
       "        -4.6849017],\n",
       "       [-2.3885746, -0.86989  , -2.1702073, ..., -2.147847 , -1.6390084,\n",
       "        -2.1693165]], dtype=float32)"
      ]
     },
     "execution_count": 65,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "preds"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "id": "f27df086-96a4-40b2-8821-5f5e58b3eacb",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T04:14:58.813201Z",
     "iopub.status.busy": "2022-12-20T04:14:58.813012Z",
     "iopub.status.idle": "2022-12-20T04:15:03.000271Z",
     "shell.execute_reply": "2022-12-20T04:15:02.999804Z",
     "shell.execute_reply.started": "2022-12-20T04:14:58.813183Z"
    },
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0., 0., 0., ..., 0., 0., 0.],\n",
       "       [0., 0., 0., ..., 0., 0., 0.],\n",
       "       [0., 0., 0., ..., 0., 0., 0.],\n",
       "       ...,\n",
       "       [0., 0., 0., ..., 0., 0., 0.],\n",
       "       [0., 0., 0., ..., 0., 0., 0.],\n",
       "       [0., 1., 0., ..., 0., 0., 0.]], dtype=float32)"
      ]
     },
     "execution_count": 66,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "trues"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "id": "b9422466-9a8e-4f8b-a574-df2db945bbfc",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T04:15:03.633156Z",
     "iopub.status.busy": "2022-12-20T04:15:03.632972Z",
     "iopub.status.idle": "2022-12-20T04:15:07.436121Z",
     "shell.execute_reply": "2022-12-20T04:15:07.435641Z",
     "shell.execute_reply.started": "2022-12-20T04:15:03.633139Z"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "from sklearn.metrics import roc_auc_score, average_precision_score, mean_squared_error\n",
    "from scipy import stats\n",
    "def evaluate_model(y_test, pred, task, verbose=True):\n",
    "    if task == 'regression': #isinstance(pl_model.criterion, torch.nn.modules.loss.MSELoss):\n",
    "        mse = calculate_mse(y_test, pred)\n",
    "        pearsonr = calculate_pearsonr(y_test, pred)\n",
    "        spearmanr = calculate_spearmanr(y_test, pred)\n",
    "        if verbose:\n",
    "            print(\"Test MSE       : %.4f +/- %.4f\"%(np.nanmean(mse), np.nanstd(mse)))\n",
    "            print(\"Test Pearson r : %.4f +/- %.4f\"%(np.nanmean(pearsonr), np.nanstd(pearsonr)))\n",
    "            print(\"Test Spearman r: %.4f +/- %.4f\"%(np.nanmean(spearmanr), np.nanstd(spearmanr)))\n",
    "        return mse, pearsonr, spearmanr\n",
    "\n",
    "    else: \n",
    "        auroc = calculate_auroc(y_test, pred) \n",
    "        aupr = calculate_aupr(y_test, pred) \n",
    "        if verbose:\n",
    "            print(\"Test AUROC: %.4f +/- %.4f\"%(np.nanmean(auroc), np.nanstd(auroc)))\n",
    "            print(\"Test AUPR : %.4f +/- %.4f\"%(np.nanmean(aupr), np.nanstd(aupr)))\n",
    "        return auroc, aupr\n",
    "\n",
    "\n",
    "def calculate_auroc(y_true, y_score):\n",
    "    vals = []\n",
    "    for class_index in range(y_true.shape[-1]):\n",
    "        vals.append( roc_auc_score(y_true[:,class_index], y_score[:,class_index]) )    \n",
    "    return np.array(vals)\n",
    "\n",
    "def calculate_aupr(y_true, y_score):\n",
    "    vals = []\n",
    "    for class_index in range(y_true.shape[-1]):\n",
    "        vals.append( average_precision_score(y_true[:,class_index], y_score[:,class_index]) )    \n",
    "    return np.array(vals)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "id": "6bbdb27f-3299-4323-ace9-75ec5bc3b0c3",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T04:15:07.437090Z",
     "iopub.status.busy": "2022-12-20T04:15:07.436911Z",
     "iopub.status.idle": "2022-12-20T04:15:13.020012Z",
     "shell.execute_reply": "2022-12-20T04:15:13.019493Z",
     "shell.execute_reply.started": "2022-12-20T04:15:07.437073Z"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "results = evaluate_model(trues, preds, \"binary\", verbose=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "id": "f90db022-f08f-4664-8299-ce18751965a8",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T04:15:39.590946Z",
     "iopub.status.busy": "2022-12-20T04:15:39.590627Z",
     "iopub.status.idle": "2022-12-20T04:15:43.542884Z",
     "shell.execute_reply": "2022-12-20T04:15:43.542211Z",
     "shell.execute_reply.started": "2022-12-20T04:15:39.590916Z"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "aurocs = results[0]\n",
    "auprs = results[1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "id": "e0ea2bcd-d938-4ccb-b8de-8701085e22fd",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T04:15:55.042529Z",
     "iopub.status.busy": "2022-12-20T04:15:55.042288Z",
     "iopub.status.idle": "2022-12-20T04:15:59.000309Z",
     "shell.execute_reply": "2022-12-20T04:15:58.999673Z",
     "shell.execute_reply.started": "2022-12-20T04:15:55.042511Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(0.8451581652547495, 0.5309917392165966)"
      ]
     },
     "execution_count": 71,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "aurocs.mean(), auprs.mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "id": "541db2b2-b8d8-4ce5-aeab-2f2c373df6b0",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T04:18:46.971320Z",
     "iopub.status.busy": "2022-12-20T04:18:46.971122Z",
     "iopub.status.idle": "2022-12-20T04:19:19.085919Z",
     "shell.execute_reply": "2022-12-20T04:19:19.085410Z",
     "shell.execute_reply.started": "2022-12-20T04:18:46.971302Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Already up to date.\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "2    Basset\n",
       "Name: model, dtype: object"
      ]
     },
     "execution_count": 72,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "eu.external.kipoi.get_model_names(\"Basset\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "id": "3a084529-9564-4955-ba5b-b90aa0d886d6",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T04:20:26.886526Z",
     "iopub.status.busy": "2022-12-20T04:20:26.886141Z",
     "iopub.status.idle": "2022-12-20T04:20:31.715671Z",
     "shell.execute_reply": "2022-12-20T04:20:31.715011Z",
     "shell.execute_reply.started": "2022-12-20T04:20:26.886505Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Using downloaded and verified file: /cellar/users/aklie/.kipoi/models/Basset/downloaded/model_files/weights/4878981d84499eb575abd0f3b45570d3\n"
     ]
    }
   ],
   "source": [
    "kipoi_model = eu.external.kipoi.get_model(\"Basset\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "id": "8cef8bf6-dcc4-4d93-8f5a-61a0bf064584",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T04:21:38.591625Z",
     "iopub.status.busy": "2022-12-20T04:21:38.591452Z",
     "iopub.status.idle": "2022-12-20T04:21:42.589236Z",
     "shell.execute_reply": "2022-12-20T04:21:42.588708Z",
     "shell.execute_reply.started": "2022-12-20T04:21:38.591608Z"
    }
   },
   "outputs": [],
   "source": [
    "import torch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 100,
   "id": "d1f2eece-4c9a-4af1-bbb9-e5980fa68ae7",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T04:28:58.773159Z",
     "iopub.status.busy": "2022-12-20T04:28:58.772987Z",
     "iopub.status.idle": "2022-12-20T04:29:02.646425Z",
     "shell.execute_reply": "2022-12-20T04:29:02.645828Z",
     "shell.execute_reply.started": "2022-12-20T04:28:58.773142Z"
    },
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([1, 4, 600, 1])"
      ]
     },
     "execution_count": 100,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "torch.Tensor(np.expand_dims(data_module.x_test, axis=3))[:1].size()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 111,
   "id": "5394ed02-3148-4392-a438-b7626cea17e7",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T04:31:26.737241Z",
     "iopub.status.busy": "2022-12-20T04:31:26.736930Z",
     "iopub.status.idle": "2022-12-20T04:31:30.785340Z",
     "shell.execute_reply": "2022-12-20T04:31:30.784798Z",
     "shell.execute_reply.started": "2022-12-20T04:31:26.737223Z"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "kipoi_preds = kipoi_model(torch.Tensor(np.expand_dims(data_module.x_test, axis=3))[:1000].to(\"cuda\")).to(\"cpu\").detach().numpy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 112,
   "id": "e9c6093f-e819-4291-bbf4-c293079b0601",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T04:31:30.786397Z",
     "iopub.status.busy": "2022-12-20T04:31:30.786218Z",
     "iopub.status.idle": "2022-12-20T04:31:35.049564Z",
     "shell.execute_reply": "2022-12-20T04:31:35.049050Z",
     "shell.execute_reply.started": "2022-12-20T04:31:30.786381Z"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "results = evaluate_model(trues[:1000], kipoi_preds, \"binary\", verbose=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 114,
   "id": "6a82ece7-9263-411f-9119-b4a6c0d06457",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T04:31:55.223578Z",
     "iopub.status.busy": "2022-12-20T04:31:55.223339Z",
     "iopub.status.idle": "2022-12-20T04:31:59.259702Z",
     "shell.execute_reply": "2022-12-20T04:31:59.259228Z",
     "shell.execute_reply.started": "2022-12-20T04:31:55.223560Z"
    },
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.6093829996690903"
      ]
     },
     "execution_count": 114,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "results[1].mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 116,
   "id": "2e7d0389-df58-475f-ad7b-6ffe75b05f26",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T04:32:26.591590Z",
     "iopub.status.busy": "2022-12-20T04:32:26.591345Z",
     "iopub.status.idle": "2022-12-20T04:32:30.625696Z",
     "shell.execute_reply": "2022-12-20T04:32:30.625084Z",
     "shell.execute_reply.started": "2022-12-20T04:32:26.591571Z"
    },
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.nn.modules.container.Sequential"
      ]
     },
     "execution_count": 116,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "type(kipoi_model)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 117,
   "id": "e1088e81-db89-494b-9802-c12d69af2930",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-12-20T04:32:57.887153Z",
     "iopub.status.busy": "2022-12-20T04:32:57.886831Z",
     "iopub.status.idle": "2022-12-20T04:33:01.824834Z",
     "shell.execute_reply": "2022-12-20T04:33:01.824177Z",
     "shell.execute_reply.started": "2022-12-20T04:32:57.887136Z"
    },
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Sequential(\n",
       "  (0): Conv2d(4, 300, kernel_size=(19, 1), stride=(1, 1))\n",
       "  (1): BatchNorm2d(300, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "  (2): ReLU()\n",
       "  (3): MaxPool2d(kernel_size=(3, 1), stride=(3, 1), padding=0, dilation=1, ceil_mode=False)\n",
       "  (4): Conv2d(300, 200, kernel_size=(11, 1), stride=(1, 1))\n",
       "  (5): BatchNorm2d(200, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "  (6): ReLU()\n",
       "  (7): MaxPool2d(kernel_size=(4, 1), stride=(4, 1), padding=0, dilation=1, ceil_mode=False)\n",
       "  (8): Conv2d(200, 200, kernel_size=(7, 1), stride=(1, 1))\n",
       "  (9): BatchNorm2d(200, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "  (10): ReLU()\n",
       "  (11): MaxPool2d(kernel_size=(4, 1), stride=(4, 1), padding=0, dilation=1, ceil_mode=False)\n",
       "  (12): Lambda()\n",
       "  (13): Sequential(\n",
       "    (0): Lambda()\n",
       "    (1): Linear(in_features=2000, out_features=1000, bias=True)\n",
       "  )\n",
       "  (14): BatchNorm1d(1000, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "  (15): ReLU()\n",
       "  (16): Dropout(p=0.3, inplace=False)\n",
       "  (17): Sequential(\n",
       "    (0): Lambda()\n",
       "    (1): Linear(in_features=1000, out_features=1000, bias=True)\n",
       "  )\n",
       "  (18): BatchNorm1d(1000, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "  (19): ReLU()\n",
       "  (20): Dropout(p=0.3, inplace=False)\n",
       "  (21): Sequential(\n",
       "    (0): Lambda()\n",
       "    (1): Linear(in_features=1000, out_features=164, bias=True)\n",
       "  )\n",
       "  (22): Sigmoid()\n",
       ")"
      ]
     },
     "execution_count": 117,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "kipoi_model.modules"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "68fcc981-da9e-47f4-9be3-52446edee1cb",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.9 ml4gland",
   "language": "python",
   "name": "ml4gland"
  },
  "vscode": {
   "interpreter": {
    "hash": "f0aab14ae665ca4264878e5867720697752ca4d3a67458798aa51c276bf829a9"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}