check_scores_fava_annotated.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import pickle as pk\n",
    "\n",
    "import numpy as np\n",
    "from matplotlib import pyplot as plt\n",
    "from sklearn.metrics import auc, roc_curve\n",
    "\n",
    "\n",
    "os.environ[\"KMP_DUPLICATE_LIB_OK\"]=\"TRUE\"\n",
    "os.environ['HF_HOME'] = \"/path/to/hf_home\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/fs/nexus-scratch/sbharti/anaconda3/envs/hallu/lib/python3.9/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
      "  from .autonotebook import tqdm as notebook_tqdm\n"
     ]
    }
   ],
   "source": [
    "# load train and test scores from /data\n",
    "with open('data/scores_fava_annot_llama_500samp.pkl', 'rb') as f:\n",
    "    scores, sample_indiv_scores, sample_labels = pk.load(f)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_roc_auc_scores(scores: np.array, labels: np.array):\n",
    "    \"\"\"\n",
    "    Calculate ROC AUC metrics, including the AUC score, accuracy, and other key values.\n",
    "\n",
    "    Args:\n",
    "        scores (np.array): Predicted scores.\n",
    "        labels (np.array): Ground Truth binary labels.\n",
    "\n",
    "    Returns:\n",
    "        tuple: A tuple containing the following:\n",
    "            - arc (float): Area Under the Curve (AUC) for the Receiver Operating Characteristic (ROC) curve.\n",
    "            - acc (float): Maximum accuracy derived from the ROC curve.\n",
    "            - low (float): True Positive Rate (TPR) at the maximum False Positive Rate (FPR) < 0.05.\n",
    "            - fpr (np.ndarray): False Positive Rates at various thresholds.\n",
    "            - tpr (np.ndarray): True Positive Rates at various thresholds.\n",
    "            - thresh_ind (int): Index of the threshold corresponding to maximum accuracy.\n",
    "            - thresh (np.ndarray): Thresholds used for calculating the ROC curve.\n",
    "    \"\"\"\n",
    "    fpr, tpr, thresh = roc_curve(labels, scores)\n",
    "    arc = auc(fpr, tpr)\n",
    "    acc = np.max(1 - (fpr + (1 - tpr))/2)\n",
    "    thresh_ind = np.argmax(1 - (fpr + (1 - tpr))/2)\n",
    "    low = tpr[np.where(fpr<0.05)[0][-1]]\n",
    "    return arc, acc, low, fpr, tpr, thresh_ind, thresh\n",
    "\n",
    "\n",
    "def get_thresh_val(thresh: np.array, acc: float, scores: np.array):\n",
    "    \"\"\"Find approx threshold that matches avg accuracy\"\"\"\n",
    "    for t in thresh:\n",
    "        pred_list = np.array([ 1 if x < t else 0 for x in scores])\n",
    "        if np.mean(pred_list) <= acc:\n",
    "            #print(f\"Accuracy: {acc:.2f} , {np.mean(pred_list):.2f}\")\n",
    "            return t, pred_list\n",
    "\n",
    "def get_balanced_scores(scores: np.array, sample_labels: np.array):\n",
    "    \"\"\"Get balanced scores\"\"\"\n",
    "    num_samp = min(sum(sample_labels), len(sample_labels)-sum(sample_labels))\n",
    "    bal_sc = np.concatenate([scores[:num_samp] , scores[-num_samp:]])\n",
    "    bal_labels = np.concatenate([sample_labels[:num_samp], sample_labels[-num_samp:]])\n",
    "    return bal_sc, bal_labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(334, 334)"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Ensure balanced dataset!\n",
    "bal_sc, bal_labels = get_balanced_scores(scores,sample_labels)\n",
    "len(bal_labels), sum(bal_labels)*2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "PPL & 53.22 & 58.68 & 3.59 \\\\\n",
      "Window Entropy & 56.92 & 56.29 & 2.99 \\\\\n",
      "Logit Entropy & 53.77 & 56.29 & 2.99 \\\\\n"
     ]
    }
   ],
   "source": [
    "ly_scores = -np.array(sample_indiv_scores['logit'][\"perplexity\"])\n",
    "arc, acc, low, fpr, tpr, thresh_ind, thresh = get_roc_auc_scores(*get_balanced_scores(ly_scores,sample_labels))\n",
    "# print(f\"AUROC:{arc*100:.2f}, Acc:{acc*100:.2f}, TPR@5%FPR:{low*100:.2f}\")\n",
    "print(f\"PPL & {arc*100:.2f} & {acc*100:.2f} & {low*100:.2f} \\\\\\\\\")\n",
    "\n",
    "ly_scores = np.array(sample_indiv_scores['logit'][\"window_entropy\"])\n",
    "arc, acc, low, fpr, tpr, thresh_ind, thresh = get_roc_auc_scores(*get_balanced_scores(ly_scores,sample_labels))\n",
    "print(f\"Window Entropy & {arc*100:.2f} & {acc*100:.2f} & {low*100:.2f} \\\\\\\\\")\n",
    "\n",
    "ly_scores = np.array(sample_indiv_scores['logit'][\"logit_entropy\"])\n",
    "arc, acc, low, fpr, tpr, thresh_ind, thresh = get_roc_auc_scores(*get_balanced_scores(ly_scores,sample_labels))\n",
    "print(f\"Logit Entropy & {arc*100:.2f} & {acc*100:.2f} & {low*100:.2f} \\\\\\\\\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Layer:2 - AUROC:0.5951, Acc:0.6018, TPR@5%FPR:0.1138\n",
      "Layer:3 - AUROC:0.5889, Acc:0.5898, TPR@5%FPR:0.1018\n",
      "Layer:4 - AUROC:0.6120, Acc:0.5958, TPR@5%FPR:0.1198\n",
      "Layer:5 - AUROC:0.6449, Acc:0.6287, TPR@5%FPR:0.1557\n",
      "Layer:6 - AUROC:0.5611, Acc:0.5928, TPR@5%FPR:0.0240\n",
      "Layer:7 - AUROC:0.5653, Acc:0.5719, TPR@5%FPR:0.0419\n",
      "Layer:8 - AUROC:0.5877, Acc:0.5958, TPR@5%FPR:0.0599\n",
      "Layer:9 - AUROC:0.5693, Acc:0.5988, TPR@5%FPR:0.0659\n",
      "Layer:10 - AUROC:0.5750, Acc:0.5958, TPR@5%FPR:0.0599\n",
      "Layer:11 - AUROC:0.5569, Acc:0.5868, TPR@5%FPR:0.0719\n",
      "Layer:12 - AUROC:0.5283, Acc:0.5599, TPR@5%FPR:0.0599\n",
      "Layer:13 - AUROC:0.5466, Acc:0.5689, TPR@5%FPR:0.0359\n",
      "Layer:14 - AUROC:0.5751, Acc:0.6048, TPR@5%FPR:0.0659\n",
      "Layer:15 - AUROC:0.5416, Acc:0.5808, TPR@5%FPR:0.0180\n",
      "Layer:16 - AUROC:0.5571, Acc:0.5838, TPR@5%FPR:0.0539\n",
      "Layer:17 - AUROC:0.5954, Acc:0.6168, TPR@5%FPR:0.0539\n",
      "Layer:18 - AUROC:0.6288, Acc:0.6377, TPR@5%FPR:0.0659\n",
      "Layer:19 - AUROC:0.6995, Acc:0.6737, TPR@5%FPR:0.2335\n",
      "Layer:20 - AUROC:0.6675, Acc:0.6617, TPR@5%FPR:0.1617\n",
      "Layer:21 - AUROC:0.7237, Acc:0.6796, TPR@5%FPR:0.1497\n",
      "Layer:22 - AUROC:0.6823, Acc:0.6527, TPR@5%FPR:0.1617\n",
      "Layer:23 - AUROC:0.6904, Acc:0.6497, TPR@5%FPR:0.2156\n",
      "Layer:24 - AUROC:0.6079, Acc:0.6257, TPR@5%FPR:0.0539\n",
      "Layer:25 - AUROC:0.5983, Acc:0.5808, TPR@5%FPR:0.1257\n",
      "Layer:26 - AUROC:0.5923, Acc:0.5838, TPR@5%FPR:0.1198\n",
      "Layer:27 - AUROC:0.6261, Acc:0.6138, TPR@5%FPR:0.1078\n",
      "Layer:28 - AUROC:0.5733, Acc:0.5958, TPR@5%FPR:0.0838\n",
      "Layer:29 - AUROC:0.5790, Acc:0.6168, TPR@5%FPR:0.0719\n",
      "Layer:30 - AUROC:0.5613, Acc:0.5988, TPR@5%FPR:0.0659\n",
      "Layer:31 - AUROC:0.5169, Acc:0.5659, TPR@5%FPR:0.0299\n",
      "Layer:32 - AUROC:0.5223, Acc:0.5569, TPR@5%FPR:0.0299\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "num_layers = len(sample_indiv_scores['attns'].keys())  # The 7B and 8B models that are being evaluated have 32 layers, so num_layers=31\n",
    "arc_list, acc_list, low_list = [], [], []\n",
    "\n",
    "samp_preds = []\n",
    "thresh_vals = []\n",
    "\n",
    "for layer_num in range(1, num_layers + 1):\n",
    "    scores = -np.array(sample_indiv_scores['attns'][\"Attn\"+str(layer_num)])\n",
    "    bal_sc, bal_labels = get_balanced_scores(scores,sample_labels)\n",
    "    arc, acc, low, fpr, tpr, thresh_ind, thresh = get_roc_auc_scores(bal_sc,bal_labels)\n",
    "    thresh_val, pred_list = get_thresh_val(thresh, acc, bal_sc)\n",
    "    samp_preds.append(pred_list)\n",
    "    thresh_vals.append(thresh_val)\n",
    "    print(f\"Layer:{layer_num+1} - AUROC:{arc:.4f}, Acc:{acc:.4f}, TPR@5%FPR:{low:.4f}\")\n",
    "    arc_list.append(arc)\n",
    "    acc_list.append(acc)\n",
    "    low_list.append(low)\n",
    "    plt.plot(fpr, tpr, label = f'LY{layer_num}, AUC={arc*100:.2f}, Acc={acc*100:.2f}, TPR@5%FPR={low*100:.2f}') \n",
    "\n",
    "# print(\"Layer-wise results\")\n",
    "# print(arc_list)\n",
    "# print(acc_list)\n",
    "# print(low_list)\n",
    "\n",
    "# # add the legend below the plot\n",
    "# plt.legend(loc='upper center', bbox_to_anchor=(0.5, -0.05), shadow=True, ncol=2)\n",
    "# plt.ylabel('True Positive Rate')\n",
    "# plt.xlabel('False Positive Rate')\n",
    "# plt.savefig(\"plots/llama_attns_layerwise_roc_var.png\", dpi=200, bbox_inches='tight')\n",
    "\n",
    "# clear the plot\n",
    "plt.clf()\n",
    "plt.xticks([1, 5, 10, 15, 20, 25, 30, 32])\n",
    "plt.plot(range(1, len(arc_list) + 1), arc_list, label='AUROC')\n",
    "plt.plot(range(1, len(arc_list) + 1), acc_list, label='ACC')\n",
    "plt.plot(range(1, len(arc_list) + 1), low_list, label='TPR@5%FPR')\n",
    "plt.xlabel('Layer Number')\n",
    "plt.grid()\n",
    "plt.legend()\n",
    "plt.savefig(\"plots/llama_hidden_fava.png\", dpi=200, bbox_inches='tight')\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "hallu",
   "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": "3.9.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}