token-level.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "bcb3e880",
   "metadata": {},
   "source": [
    "# Token Level "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "78727e84",
   "metadata": {},
   "source": [
    "## Import Dependencies  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "9e08951d",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2022-10-09 00:39:07.101922: W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libcudart.so.11.0'; dlerror: libcudart.so.11.0: cannot open shared object file: No such file or directory\n",
      "2022-10-09 00:39:07.101944: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "import os \n",
    "from itertools import repeat \n",
    "from transformers import AutoTokenizer, AutoModelForTokenClassification, pipeline\n",
    "from cleanlab.rank import get_label_quality_scores \n",
    "from utils import readfile, get_probs, get_pred_probs \n",
    "\n",
    "from cleanlab.internal.token_classification_utils import get_sentence, filter_sentence\n",
    "import matplotlib.pyplot as plt \n",
    "os.environ[\"TOKENIZERS_PARALLELISM\"] = \"false\" \n",
    "from sklearn import metrics "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "728cdceb",
   "metadata": {},
   "source": [
    "## Read datasets: original CoNLL-2003 & CoNLL++ "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "7d4174b8",
   "metadata": {},
   "outputs": [],
   "source": [
    "filepath = 'data/conll.txt'\n",
    "given_words, given_labels = readfile(filepath) \n",
    "sentences = list(map(get_sentence, given_words)) \n",
    "\n",
    "sentences, mask = filter_sentence(sentences) \n",
    "given_words = [words for m, words in zip(mask, given_words) if m] \n",
    "given_labels = [labels for m, labels in zip(mask, given_labels) if m] \n",
    "labels = [label for given_label in given_labels for label in given_label] "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "4c73603e",
   "metadata": {},
   "outputs": [],
   "source": [
    "ground_truth_path = 'data/conllpp.txt' \n",
    "ground_truth_labels = readfile(ground_truth_path)[1] \n",
    "ground_truth_labels = [label for m, label in zip(mask, ground_truth_labels) if m] \n",
    "ground_truth_labels = [label for labels in ground_truth_labels for label in labels] "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5e91efe5",
   "metadata": {},
   "source": [
    "## Distribution of mislabels \n",
    "\n",
    "The cell in the $i$th row and $j$th column below lists the number (or percentage) of tokens that are labeled class i in CoNLL++ (ground truth label) and labeled as class j in the original CoNLL-2003 dataset. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "8a1b863a",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tO\tB-MISC\tI-MISC\tB-PER\tI-PER\tB-ORG\tI-ORG\tB-LOC\tI-LOC\n",
      "O\t38132\t3\t6\t2\t0\t3\t0\t1\t0\n",
      "B-MISC\t39\t665\t0\t0\t0\t1\t0\t18\t0\n",
      "I-MISC\t48\t4\t202\t0\t0\t0\t0\t0\t0\n",
      "B-PER\t8\t2\t0\t1603\t0\t1\t0\t3\t0\n",
      "I-PER\t2\t0\t0\t5\t1153\t0\t0\t0\t1\n",
      "B-ORG\t14\t20\t0\t3\t0\t1644\t0\t32\t0\n",
      "I-ORG\t28\t0\t6\t0\t3\t3\t831\t2\t7\n",
      "B-LOC\t15\t8\t0\t3\t0\t7\t1\t1612\t0\n",
      "I-LOC\t7\t0\t2\t0\t0\t0\t1\t0\t249\n"
     ]
    }
   ],
   "source": [
    "cm = metrics.confusion_matrix(ground_truth_labels, labels) \n",
    "entities = ['O', 'B-MISC', 'I-MISC', 'B-PER', 'I-PER', 'B-ORG', 'I-ORG', 'B-LOC', 'I-LOC'] \n",
    "\n",
    "print('\\t' + '\\t'.join(entities)) \n",
    "for entity, dist in zip(entities, cm): \n",
    "    s = [str(d) for d in dist] \n",
    "    print(entity + '\\t' + '\\t'.join(s)) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "05290d71",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tO\tB-MISC\tI-MISC\tB-PER\tI-PER\tB-ORG\tI-ORG\tB-LOC\tI-LOC\n",
      "O\t99.96%\t 0.01%\t 0.02%\t 0.01%\t 0.00%\t 0.01%\t 0.00%\t 0.00%\t 0.00%\n",
      "B-MISC\t 5.39%\t91.98%\t 0.00%\t 0.00%\t 0.00%\t 0.14%\t 0.00%\t 2.49%\t 0.00%\n",
      "I-MISC\t18.90%\t 1.57%\t79.53%\t 0.00%\t 0.00%\t 0.00%\t 0.00%\t 0.00%\t 0.00%\n",
      "B-PER\t 0.49%\t 0.12%\t 0.00%\t99.13%\t 0.00%\t 0.06%\t 0.00%\t 0.19%\t 0.00%\n",
      "I-PER\t 0.17%\t 0.00%\t 0.00%\t 0.43%\t99.31%\t 0.00%\t 0.00%\t 0.00%\t 0.09%\n",
      "B-ORG\t 0.82%\t 1.17%\t 0.00%\t 0.18%\t 0.00%\t95.97%\t 0.00%\t 1.87%\t 0.00%\n",
      "I-ORG\t 3.18%\t 0.00%\t 0.68%\t 0.00%\t 0.34%\t 0.34%\t94.43%\t 0.23%\t 0.80%\n",
      "B-LOC\t 0.91%\t 0.49%\t 0.00%\t 0.18%\t 0.00%\t 0.43%\t 0.06%\t97.93%\t 0.00%\n",
      "I-LOC\t 2.70%\t 0.00%\t 0.77%\t 0.00%\t 0.00%\t 0.00%\t 0.39%\t 0.00%\t96.14%\n"
     ]
    }
   ],
   "source": [
    "percentage = np.sum(cm, axis=1)[:, np.newaxis] \n",
    "percentage = cm / percentage * 100 \n",
    "percentage = np.round(percentage, 2) \n",
    "\n",
    "def reformat(val): \n",
    "    val = str(val) \n",
    "    val = val + '%' if val.index('.') == len(val)-3 else val + '0%' \n",
    "    val = ' ' * (6-len(val)) + val \n",
    "    return val \n",
    "\n",
    "print('\\t' + '\\t'.join(entities)) \n",
    "for entity, per in zip(entities, percentage): \n",
    "    s = [reformat(p) for p in per] \n",
    "    print(entity + '\\t' + '\\t'.join(s)) "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "46d9bcdc",
   "metadata": {},
   "source": [
    "## Token-level label error detection "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "d3c34d9c",
   "metadata": {},
   "outputs": [],
   "source": [
    "model_name = 'dslim/bert-base-NER'\n",
    "model = AutoModelForTokenClassification.from_pretrained(model_name) \n",
    "tokenizer = AutoTokenizer.from_pretrained(model_name)\n",
    "pipe = pipeline(task=\"ner\", model=model, tokenizer=tokenizer)\n",
    "\n",
    "sentence_tokens = [[tokenizer.decode(token) for token in tokenizer(sentence)['input_ids']] for sentence in sentences] \n",
    "sentence_tokens = [[token.replace('#', '') for token in sentence_token][1:-1] for sentence_token in sentence_tokens] \n",
    "\n",
    "errors = np.array(ground_truth_labels) != np.array(labels) \n",
    "n_error = sum(errors) \n",
    "error_percentage = n_error / len(errors) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "3a09a91a",
   "metadata": {},
   "outputs": [],
   "source": [
    "sentence_probs = list(map(get_probs, repeat(pipe), sentences))\n",
    "pred_probs = list(map(get_pred_probs, sentence_probs, sentence_tokens, given_words)) \n",
    "pred_probs = [pred for pred_prob in pred_probs for pred in pred_prob] \n",
    "pred_probs = np.array(pred_probs) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "b6d16a9e",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                              AUPRC     AUPRC     Lift @ Error\n",
      "self_confidence               0.3483    0.9545    62.4867   \n",
      "normalized_margin             0.3296    0.9552    66.3618   \n",
      "confidence_weighted_entropy   0.2293    0.9355    46.9861   \n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "methods = [\"self_confidence\", \"normalized_margin\", \"confidence_weighted_entropy\"] \n",
    "colors = ['blue', 'red', 'green'] \n",
    "\n",
    "print('{:30}{:10}{:10}{:10}'.format('', 'AUPRC', 'AUPRC', 'Lift @ Error')) \n",
    "plt.figure(figsize=(10, 6)) \n",
    "for color, method in zip(colors, methods): \n",
    "    scores = get_label_quality_scores(labels, pred_probs, method=method) \n",
    "    ranking = np.argsort(scores) \n",
    "    error_retrival = [errors[r] for r in ranking] \n",
    "    \n",
    "    error_counts = [] \n",
    "    for e in error_retrival: \n",
    "        if len(error_counts) == 0: \n",
    "            error_counts.append(int(e)) \n",
    "        else: \n",
    "            error_counts.append(error_counts[-1] + int(e)) \n",
    "\n",
    "    precision = error_counts / (np.arange(len(error_retrival)) + 1) \n",
    "    recall = error_counts / n_error \n",
    "    \n",
    "    fpr, tpr, thresholds = metrics.roc_curve(errors, -scores) \n",
    "    auroc = np.round(metrics.auc(fpr, tpr), 4) \n",
    "    auprc = np.round(metrics.auc(recall, precision), 4) \n",
    "    lift = np.round(precision[n_error] / error_percentage, 4) \n",
    "    \n",
    "    print('{:30}{:<10}{:<10}{:<10}'.format(method, auprc, auroc, lift)) \n",
    "    \n",
    "    plt.plot(recall, precision, color=color) \n",
    "    \n",
    "plt.legend(methods) \n",
    "plt.xlabel('Recall') \n",
    "plt.ylabel('Precision') \n",
    "plt.savefig('token-level.png', dpi=400) \n",
    "plt.show() "
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.9.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}