agnews_pipeline.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import time\n",
    "import json\n",
    "from copy import deepcopy as dc\n",
    "import torch\n",
    "\n",
    "import joblib\n",
    "import re\n",
    "\n",
    "%load_ext autoreload\n",
    "%autoreload 2\n",
    "\n",
    "import matplotlib\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "\n",
    "from data.utils import votes_filter, plot_overlap_conflict\n",
    "from data.agnews.utils import load_data, get_inst"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "label_idx = [0,1,2,3]\n",
    "\n",
    "\n",
    "train = load_data('train')\n",
    "dev = load_data('dev')\n",
    "test = load_data('test')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "from nplm.plf import WeakRule, BinaryRERules, regex_decision\n",
    "\n",
    "train_instances, _ = get_inst(train)\n",
    "dev_instances, dev_labels = get_inst(dev)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.7142857142857143 0.7142857142857143 0.03\n",
      "[0.0, 1.0, 0.0, 0.0]\n",
      "0.8333333333333334 0.8333333333333334 0.01\n",
      "[1.0, nan, nan, 0.0]\n",
      "0.8888888888888888 0.8888888888888888 0.016\n",
      "[0.0, 1.0, nan, nan]\n",
      "0.9230769230769231 0.9230769230769231 0.024\n",
      "[nan, nan, 1.0, 0.0]\n",
      "0.9523809523809523 0.9523809523809523 0.04\n",
      "[nan, nan, 0.0, 1.0]\n",
      "1.0 1.0 0.044\n",
      "[1.0, 1.0, nan, nan]\n",
      "1.0 1.0 0.014\n",
      "[nan, nan, 1.0, 1.0]\n",
      "0.9642857142857143 0.9642857142857143 0.054\n",
      "[1.0, nan, 1.0, 0.0]\n",
      "1.0 1.0 0.03\n",
      "[1.0, nan, nan, 1.0]\n"
     ]
    }
   ],
   "source": [
    "def df1(instance):\n",
    "    entity_doc = instance['sen_ner']\n",
    "    if 'EVENT' in  entity_doc and 'GPE' in entity_doc:\n",
    "        return 1\n",
    "    return -1\n",
    "\n",
    "gpe_event_title = WeakRule(name='gpe+event_title', exec_module=df1, label_maps={0:[0,2,3], 1:[1]})\n",
    "\n",
    "\n",
    "v1 = gpe_event_title.execute(dev_instances)\n",
    "acc, p, r, cwacc = gpe_event_title.eval(dev_labels, class_wise_acc=True)\n",
    "print(acc, p, r)\n",
    "print(cwacc)\n",
    "\n",
    "\n",
    "def df2(instance):\n",
    "    entity_doc = instance['title_ner']\n",
    "    if 'LOC' in entity_doc:\n",
    "        return 1\n",
    "    return -1\n",
    "\n",
    "loc_title_rule = WeakRule(name='loc_title', exec_module=df2, label_maps={0:[1,2,3], 1:[0]})\n",
    "\n",
    "v1 = loc_title_rule.execute(dev_instances)\n",
    "acc, p, r, cwacc = loc_title_rule.eval(dev_labels, class_wise_acc=True)\n",
    "print(acc, p, r)\n",
    "print(cwacc)\n",
    "\n",
    "\n",
    "\n",
    "sports_names = 'kelvim escobar|red sox|formula one|grand prix|svetlana kuznetsova|billy wagner| kobe | beckham|johnny damon|robin ventura|olivier panis'\n",
    "\n",
    "sp0 = BinaryRERules(name='sports_names',re_pattern=sports_names, \n",
    "                                 preproc=lambda inst:inst['sen_lemma'].lower(), \n",
    "                                 label_maps={0:[0,2,3], 1:[1]}, unipolar=True)\n",
    "v5 = sp0.execute(dev_instances)\n",
    "\n",
    "acc, p, r, cwacc = sp0.eval(dev_labels, class_wise_acc=True)\n",
    "print(acc, p, r)\n",
    "print(cwacc)\n",
    "\n",
    "\n",
    "def bexclusive_lemmas(instance):\n",
    "\n",
    "    business_keywords = {'profit', 'bankrupt', 'yen', 'financial'}\n",
    "    \n",
    "    doc = instance['sen_lemma'].lower().split()\n",
    "    \n",
    "    for word in doc:\n",
    "        for keyword in business_keywords:\n",
    "            if word.startswith(keyword):\n",
    "                return 1\n",
    "    return -1\n",
    "\n",
    "\n",
    "bexclusive_lemmas_rule = WeakRule(name='bus_lemma', exec_module=bexclusive_lemmas, label_maps={0:[0,1,3], 1:[2]})\n",
    "v1 = bexclusive_lemmas_rule.execute(dev_instances)\n",
    "acc, p, r, cwacc = bexclusive_lemmas_rule.eval(dev_labels, class_wise_acc=True)\n",
    "print(acc, p, r)\n",
    "print(cwacc)\n",
    "\n",
    "\n",
    "tech = 'space.com|space station|network authentication|(python|java|matlab|c) developer|application|virus|browser hijack|search engine|internet-based|windows update|smart phone|source code|mangement software|software.*develop|internet connection|interactive gam|game console|transfer datum|internet security|g network|internet company|storage capacity|music player|microsystem|comsumer electronic|operat.*system|wireness network|motherboard|spacecraft|malicious program|video game'\n",
    "\n",
    "techr = BinaryRERules(name='tech_terms',re_pattern=tech, \n",
    "                                 preproc=lambda inst:inst['sen_lemma'].lower(), \n",
    "                                 label_maps={0:[0,1,2], 1:[3]}, unipolar=True)\n",
    "v5 = techr.execute(dev_instances)\n",
    "\n",
    "acc, p, r, cwacc = techr.eval(dev_labels, class_wise_acc=True)\n",
    "print(acc, p, r)\n",
    "print(cwacc)\n",
    "\n",
    "\n",
    "def exclusive_lemmas(instance):\n",
    "    \n",
    "    \n",
    "    world_pol_keywords = {'mideast', 'iraq', 'baghdad', 'pakistan', 'afghan',\n",
    "                          'kurd', 'arab' 'egypt', 'iran', 'turkey', 'syria', 'bahrain',\n",
    "                          'israel', 'jordan', 'kuwait', 'lebanon', 'oman', 'palestine',\n",
    "                          'qatar', 'saudi', 'uae', 'yemen' ,'chechnya'}\n",
    "    \n",
    "    \n",
    "    world_pol_keywords |= {'al-qaeda', 'taliban'}\n",
    "    world_pol_keywords |= {'hostage', 'abduct', 'hijack'}\n",
    "    world_pol_keywords |= {'invasion', ' coup ', 'curfew', 'army', 'troop', 'peace', 'militant', 'missile'}\n",
    "    world_pol_keywords |= {'murder', 'death'}\n",
    "#     world_pol_keywords |= {'typhoon'}\n",
    "    \n",
    "    sports_keywords = {'baseball', 'football', 'soccer', 'hockey', 'basketball', 'tennis', 'golf'}\n",
    "    sports_keywords |= {'stadium', 'arena'}\n",
    "    sports_keywords |= {'season', 'playoff', 'tournament'}\n",
    "    sports_keywords |= {'mlb', 'nfl', 'nba', 'mls', 'nhl', 'ncaa', 'league', 'racing'}\n",
    "    sports_keywords |= {'premiership'}\n",
    "    sports_keywords |= {'quarterback', 'centerback', 'fullback', 'pitcher'}\n",
    "    \n",
    "    business_keywords = {'profit', 'bankrupt', 'financial'}\n",
    "    \n",
    "    bt_keywords = {'robot', 'robotic'}\n",
    "    bt_keywords |= {'web', 'internet'}\n",
    "    bt_keywords |= {'linux'}\n",
    "    bt_keywords |= {'stem-cell', 'biotechnology'}\n",
    "    bt_keywords |= {'xbox', 'playstation'}\n",
    "    bt_keywords |= {'microsoft'}\n",
    "    bt_keywords |= {'space', 'nasa'}\n",
    "    bt_keywords |= {'adobe', 'ipod', 'apple', 'xerox', 'ibm'}\n",
    "    \n",
    "    \n",
    "    doc = instance['sen_lemma'].lower().split()\n",
    "    \n",
    "    for word in doc:\n",
    "        for keyword in bt_keywords:\n",
    "            if word.startswith(keyword):\n",
    "                return 3\n",
    "            \n",
    "    for word in doc:\n",
    "        for keyword in world_pol_keywords:\n",
    "            if word.startswith(keyword):\n",
    "                return 0\n",
    "\n",
    "    for word in doc:\n",
    "        for keyword in sports_keywords:\n",
    "            if word.startswith(keyword):\n",
    "                return 1\n",
    "            \n",
    "    for word in doc:\n",
    "        for keyword in business_keywords:\n",
    "            if word.startswith(keyword):\n",
    "                return 2\n",
    "            \n",
    "\n",
    "\n",
    "    \n",
    "    return -1\n",
    "\n",
    "exclusive_lemmas_rule = WeakRule(exec_module=exclusive_lemmas, label_maps={0:[0], 1:[1], 2:[2], 3:[2,3]})\n",
    "v1 = exclusive_lemmas_rule.execute(dev_instances)\n",
    "acc, p, r, cwacc = exclusive_lemmas_rule.eval(dev_labels, class_wise_acc=True)\n",
    "\n",
    "\n",
    "def df3(instance):\n",
    "    entity_doc = instance['title_ner']\n",
    "#     print(entity_doc)\n",
    "    if 'PERSON' in  entity_doc and 'GPE' in entity_doc:\n",
    "        return 1\n",
    "    return -1\n",
    "\n",
    "person_gpe_title_rule = WeakRule(name='p+gpe_title', exec_module=df3, label_maps={0:[2, 3], 1:[0, 1]})\n",
    "\n",
    "def df4(instance):\n",
    "    entity_doc = instance['sen_ner']\n",
    "#     print(entity_doc)\n",
    "    if 'PERSON' in  entity_doc and 'EVENT' in entity_doc:\n",
    "        return 1\n",
    "    return -1\n",
    "\n",
    "person_event_sen_rule = WeakRule(name='person+event_sen', exec_module=df4, label_maps={0:[2,3], 1:[0,1]})\n",
    "\n",
    "\n",
    "v1 = person_event_sen_rule.execute(dev_instances)\n",
    "acc, p, r, cwacc = person_event_sen_rule.eval(dev_labels, class_wise_acc=True)\n",
    "print(acc, p, r)\n",
    "print(cwacc)\n",
    "\n",
    "def df5(instance):\n",
    "    entity_doc = instance['title_ner']\n",
    "    if  'PRODUCT' in entity_doc:\n",
    "        return 1\n",
    "    return -1\n",
    "\n",
    "sen_prod_t_rule = WeakRule(exec_module=df5, label_maps={0:[0,1], 1:[2,3]})\n",
    "\n",
    "v1 = sen_prod_t_rule.execute(dev_instances)\n",
    "acc, p, r, cwacc = sen_prod_t_rule.eval(dev_labels, class_wise_acc=True)\n",
    "print(acc, p, r)\n",
    "print(cwacc)\n",
    "\n",
    "\n",
    "\n",
    "test_patt = 'minister|chairman'\n",
    "dpw2 = BinaryRERules(name='dpw2',\n",
    "                     re_pattern=test_patt, \n",
    "                                 preproc=lambda inst:inst['sen_lemma'].lower(), \n",
    "                                 label_maps={0:[1,3], 1:[0,2]}, unipolar=True)\n",
    "v5 = dpw2.execute(dev_instances)\n",
    "\n",
    "acc, p, r, cwacc = dpw2.eval(dev_labels, class_wise_acc=True)\n",
    "print(acc, p, r)\n",
    "print(cwacc)\n",
    "\n",
    "\n",
    "intn = 'internet'\n",
    "\n",
    "intnr = BinaryRERules(name='tech',re_pattern=intn, \n",
    "                                 preproc=lambda inst:inst['sen_lemma'].lower(), \n",
    "                                 label_maps={0:[1,2], 1:[0,3]}, unipolar=True)\n",
    "v5 = intnr.execute(dev_instances)\n",
    "\n",
    "acc, p, r, cwacc = intnr.eval(dev_labels, class_wise_acc=True)\n",
    "print(acc, p, r)\n",
    "print(cwacc)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{0: [0, 2, 3], 1: [1]}\n",
      "{0: [1, 3], 1: [0, 2]}\n",
      "{0: [1, 2], 1: [0, 3]}\n",
      "{0: [1, 2, 3], 1: [0]}\n",
      "{0: [0, 2, 3], 1: [1]}\n",
      "{0: [0], 1: [1], 2: [2], 3: [2, 3]}\n",
      "{0: [2, 3], 1: [0, 1]}\n",
      "{0: [0, 1, 3], 1: [2]}\n",
      "{0: [0, 1, 2], 1: [3]}\n",
      "{0: [2, 3], 1: [0, 1]}\n",
      "{0: [0, 1], 1: [2, 3]}\n"
     ]
    }
   ],
   "source": [
    "from nplm.plf import executor, evaluator\n",
    "\n",
    "plfs = [\n",
    "    gpe_event_title,\n",
    "    dpw2,\n",
    "    intnr,\n",
    "    loc_title_rule,\n",
    "    sp0,\n",
    "    exclusive_lemmas_rule,\n",
    "    person_gpe_title_rule,\n",
    "    bexclusive_lemmas_rule,\n",
    "    techr,\n",
    "    person_event_sen_rule,\n",
    "    sen_prod_t_rule,\n",
    "]\n",
    "\n",
    "votes, fid2clusters = executor(plfs, train_instances, one_indexed=False)\n",
    "# Filter out instances with no votes\n",
    "votes, _, train_idx = votes_filter(votes, return_idx=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_overlap_conflict(votes, fid2clusters)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "epoch   1:  12%|█▎        | 1/8 [00:00<00:01,  5.80it/s, Train Loss: =9.12]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Setup:  0.04872012138366699\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "epoch   1: 100%|██████████| 8/8 [00:01<00:00,  6.54it/s, Epoch Loss: =8.94]\n",
      "epoch   2: 100%|██████████| 8/8 [00:01<00:00,  6.63it/s, Epoch Loss: =8.53]\n",
      "epoch   3: 100%|██████████| 8/8 [00:01<00:00,  6.60it/s, Epoch Loss: =8.15]\n",
      "epoch   4: 100%|██████████| 8/8 [00:01<00:00,  6.63it/s, Epoch Loss: =7.8] \n",
      "epoch   5: 100%|██████████| 8/8 [00:01<00:00,  6.55it/s, Epoch Loss: =7.48]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Setup:  0.03786277770996094\n",
      "Parallel Estimation:  0.03423476219177246\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "from nplm import PartialLabelModel\n",
    "\n",
    "lm_cfg={'lr': 0.01,\n",
    " 'epoch': 5,\n",
    " 'seed': 0,\n",
    " 'batch_size': 8192,\n",
    " 'momentum': 0.9,\n",
    " 'step_schedule': 'p',\n",
    " 'step_multiplier': 0.1}\n",
    "\n",
    "device = 'cuda:0' if torch.cuda.is_available() else 'cpu'\n",
    "labelmodel = PartialLabelModel(num_classes=4,\n",
    "                        fid2clusters=fid2clusters,\n",
    "                        opt_cfg=lm_cfg,\n",
    "                        device=device, verbose=True)\n",
    "\n",
    "labelmodel.optimize(votes)\n",
    "label_estim = labelmodel.weak_label(votes)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "np.save('ag_news_soft_labels', {'sl': label_estim, 'idx': train_idx})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{0: [0, 2, 3], 1: [1]}\n",
      "{0: [1, 3], 1: [0, 2]}\n",
      "{0: [1, 2], 1: [0, 3]}\n",
      "{0: [1, 2, 3], 1: [0]}\n",
      "{0: [0, 2, 3], 1: [1]}\n",
      "{0: [0], 1: [1], 2: [2], 3: [2, 3]}\n",
      "{0: [2, 3], 1: [0, 1]}\n",
      "{0: [0, 1, 3], 1: [2]}\n",
      "{0: [0, 1, 2], 1: [3]}\n",
      "{0: [2, 3], 1: [0, 1]}\n",
      "{0: [0, 1], 1: [2, 3]}\n",
      "Name\tAcc\tP\tR\n",
      "gpe+event_title \t0.7163\t0.7163\t0.0329\n",
      "dpw2 \t0.9228\t0.9228\t0.0330\n",
      "tech \t0.8578\t0.8578\t0.0230\n",
      "loc_title \t0.4615\t0.4615\t0.0111\n",
      "sports_names \t0.9565\t0.9565\t0.0174\n",
      "Rule 5 \t0.9300\t0.9247\t0.3408\n",
      "p+gpe_title \t0.9331\t0.9331\t0.0349\n",
      "bus_lemma \t0.8532\t0.8532\t0.0367\n",
      "tech_terms \t0.9097\t0.9097\t0.0371\n",
      "person+event_sen \t0.9798\t0.9798\t0.0447\n",
      "Rule 10 \t0.8760\t0.8760\t0.0139\n",
      "Setup:  0.006567239761352539\n",
      "Parallel Estimation:  0.021131277084350586\n",
      "(0.5107894736842106, 0.7445193726708063, 0.5107894736842105, 0.49381807494357566)\n",
      "[0.51078947 0.66144737]\n"
     ]
    }
   ],
   "source": [
    "from eval import gen_stats, topk_results\n",
    "\n",
    "data_instance_test, data_label_test = get_inst(test)\n",
    "\n",
    "votes_test, _ = executor(plfs, data_instance_test, one_indexed=False)\n",
    "accs, ps, rs = evaluator(plfs, data_label_test, pprint=True)\n",
    "\n",
    "\n",
    "votes_test_filtered, label_filtered_test = votes_filter(votes_test, labels=data_label_test)\n",
    "\n",
    "\n",
    "wlabels_test = labelmodel.weak_label(votes_test)\n",
    "\n",
    "test_label_test = [elem+1 for elem in data_label_test]\n",
    "\n",
    "_, stats = gen_stats(wlabels_test, test_label_test)\n",
    "print(stats)\n",
    "print(topk_results(wlabels_test, test_label_test, [1,2]))\n"
   ]
  }
 ],
 "metadata": {
  "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": "3.8.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}