explore_results.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Exploration of results\n",
    "\n",
    "**[WARNING]** This notebooks requires output from `generate_dataframes.py` (saved in the `./data` folder)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from pathlib import Path\n",
    "import pandas as pd\n",
    "from collections import defaultdict\n",
    "import statistics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "%matplotlib inline\n",
    "import seaborn as sns\n",
    "sns.set()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Lemmas grouped by similarity of their most prominent senses:\n",
    "dLemmas = {\"all_lemmas\": [\"anger_NN\",\"apple_NN\",\"art_NN\", \"democracy_NN\",\n",
    "                  \"happiness_NN\", \"labour_NN\", \"machine_NN\", \"man_NN\",\n",
    "                  \"nation_NN\", \"power_NN\", \"slave_NN\", 'woman_NN'],\n",
    "           \n",
    "           \"non_industrial\": [\"anger_NN\",\"apple_NN\",\"democracy_NN\",\n",
    "                  \"happiness_NN\",\"man_NN\",\n",
    "                  \"nation_NN\",\"slave_NN\",'woman_NN'],\n",
    "           \n",
    "           \"industrial\": [\"art_NN\", \"labour_NN\", \"machine_NN\", \"power_NN\"],\n",
    "           \n",
    "           \"non_tech\": [\"anger_NN\",\"apple_NN\",\"art_NN\", \"democracy_NN\",\n",
    "                  \"happiness_NN\", \"labour_NN\", \"man_NN\",\n",
    "                  \"nation_NN\", \"slave_NN\", 'woman_NN'],\n",
    "           \n",
    "           \"tech\": [\"machine_NN\", \"power_NN\"],\n",
    "           \n",
    "           \"human\": [\"slave_NN\", 'woman_NN', \"man_NN\"],\n",
    "           \n",
    "           \"emotion\": [\"happiness_NN\", \"anger_NN\"],\n",
    "           \n",
    "           \"abstract\": [\"happiness_NN\", \"anger_NN\", \"art_NN\", \"democracy_NN\",\n",
    "                  \"labour_NN\", \"nation_NN\"],\n",
    "           \n",
    "           \"abstract_wo_emotions\": [\"art_NN\", \"democracy_NN\",\n",
    "                  \"labour_NN\", \"nation_NN\"],\n",
    "           \n",
    "           \"concrete\": [\"apple_NN\", \"machine_NN\", \"man_NN\", \"slave_NN\", 'woman_NN'],\n",
    "           \n",
    "           \"concrete_wo_machine\": [\"apple_NN\", \"man_NN\", \"slave_NN\", 'woman_NN'],\n",
    "           \n",
    "           \"man_apple_woman\": [\"apple_NN\", \"man_NN\", 'woman_NN'],\n",
    "           \n",
    "           \"apple\": [\"apple_NN\"],\n",
    "           \n",
    "           \"machine\": [\"machine_NN\"],\n",
    "           \n",
    "           \"slave\": [\"slave_NN\"],\n",
    "           \n",
    "           \"man_woman\": [\"man_NN\", \"woman_NN\"],\n",
    "           \n",
    "           \"all_wo_machine\": [\"anger_NN\",\"apple_NN\",\"art_NN\", \"democracy_NN\",\n",
    "                  \"happiness_NN\", \"labour_NN\", \"man_NN\",\n",
    "                  \"nation_NN\", \"power_NN\", \"slave_NN\", 'woman_NN'],\n",
    "           \n",
    "           \"work_related\": [\"slave_NN\", \"machine_NN\", \"labour_NN\", \"power_NN\"],\n",
    "           \n",
    "           \"non_work_related\": [\"anger_NN\",\"apple_NN\",\"art_NN\", \"democracy_NN\",\n",
    "                  \"happiness_NN\", \"man_NN\",\n",
    "                  \"nation_NN\", 'woman_NN']\n",
    "          }"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "###  Optimal time range for a Language Model\n",
    "\n",
    "The following two cells reproduce Figure 1. It finds the optimal date range for the different language models (measured by the F1-score of the positive class) givenn some headwords and using the _sense centroid_ method. Quotations are filtered in moving 100-year time windows."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "def find_optimal_daterange(results_path, lemmas, timestart, year_window, metric):\n",
    "    from sklearn.metrics import precision_recall_fscore_support\n",
    "    clf_dict = defaultdict(list)\n",
    "    results = {}\n",
    "    csv_files = results_path.glob(\"**/*.csv\")\n",
    "    for csv in csv_files:\n",
    "        current_csv = str(csv).split(\"/\")\n",
    "        current_lemma = current_csv[1]\n",
    "        current_sense = current_csv[3].split(\"~\")[0]\n",
    "        lemma_pickle = pd.read_pickle(\"data/lemma_senses_\" + current_lemma + \".pickle\")\n",
    "        sst = lemma_pickle[lemma_pickle[\"id\"] == current_sense].iloc[0][\"daterange.start\"]\n",
    "            \n",
    "        try:\n",
    "            df = pd.read_csv(csv)\n",
    "            \n",
    "            df = df[[\"label\",\"year\",\"quotation_id\",\"bert_centroid_sense_vector_bert_base_-1,-2,-3,-4_mean\",\n",
    "                     \"bert_centroid_sense_vector_bert_1850_-1,-2,-3,-4_mean\",\"bert_centroid_sense_vector_blert_base_-1,-2,-3,-4_mean\"]]\n",
    "            df = df.rename(columns={\"bert_centroid_sense_vector_bert_base_-1,-2,-3,-4_mean\": \"bert_base_sense_centroid\",\n",
    "                                    \"bert_centroid_sense_vector_bert_1850_-1,-2,-3,-4_mean\": \"bert_1850_sense_centroid\",\n",
    "                                    \"bert_centroid_sense_vector_blert_base_-1,-2,-3,-4_mean\": \"bert_1900_sense_centroid\"})\n",
    "            df = df[df[\"year\"].between(timestart, timestart+(year_window*2))]\n",
    "\n",
    "        except Exception as e:\n",
    "            continue\n",
    "        \n",
    "        if current_lemma in lemmas:\n",
    "            for col in df.columns:\n",
    "                clf_dict[col].extend(df[col])\n",
    "\n",
    "    for colname, classifications in clf_dict.items():\n",
    "        if colname not in ('label','year','quotation_id') and colname.startswith(\"bert_\"):\n",
    "            p,r = [round(x,3) for x in precision_recall_fscore_support(clf_dict['label'],classifications,average='binary',pos_label=1)[:2] if x] \n",
    "            f1 = round((2*(p*r))/(p+r),3)\n",
    "            if metric == \"recall\":\n",
    "                results[colname] = round(r,3)\n",
    "            if metric == \"precision\":\n",
    "                results[colname] = round(p,3)\n",
    "            if metric == \"fscore\":\n",
    "                results[colname] = round(f1,3)\n",
    "\n",
    "    return results"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "To obtain Figure 1 from the paper, `setting` should be `all_lemmas` and `metric` should be `fscore`.\n",
    "\n",
    "You can experiment further with other scenarios by changing the metrics and groupings of lemmas."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "setting = \"all_lemmas\" # Options: see dictionary \"dLemmas\" at the beginning of the notebook.\n",
    "metric = \"fscore\" # Options: \"precision\", \"recall\", \"fscore\" (this always corresponds to the positive class)\n",
    "\n",
    "lemmas = dLemmas[setting]\n",
    "\n",
    "path = Path('figures')\n",
    "path.mkdir(exist_ok=True)\n",
    "\n",
    "time_experiment = 2000 # Folder from which we select results\n",
    "test_daterange_start = 1760 # Quotations starting from\n",
    "year_window = 50\n",
    "bert_base = []\n",
    "bert_1850 = []\n",
    "bert_1900 = []\n",
    "time_mean = []\n",
    "for date_start in range(test_daterange_start, time_experiment - year_window,10):\n",
    "    results = find_optimal_daterange(Path('results_' + str(time_experiment)), lemmas, date_start, year_window, metric)\n",
    "    bert_base.append(results[\"bert_base_sense_centroid\"])\n",
    "    bert_1850.append(results[\"bert_1850_sense_centroid\"])\n",
    "    bert_1900.append(results[\"bert_1900_sense_centroid\"])\n",
    "    time_mean.append(date_start + year_window)\n",
    "    \n",
    "plt.plot(time_mean, bert_base)\n",
    "plt.plot(time_mean, bert_1850)\n",
    "plt.plot(time_mean, bert_1900)\n",
    "\n",
    "plt.legend([\"BERTbase\", \"BERT1850\", \"BERT1900\"], loc='lower right')\n",
    "\n",
    "plt.savefig(path / (\"plot_LM_\" + setting + \"_class1\" + metric + \"_50.png\"), dpi=300, bbox_inches=\"tight\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "###  Optimal time range for a Language Model: weighted sense centroid\n",
    "\n",
    "This does the same as the preceding two cells, but for the time-sensitive approaches."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "def find_optimal_daterange_ts(results_path, lemmas, timestart, year_window, metric, weighting):\n",
    "    from sklearn.metrics import precision_recall_fscore_support\n",
    "    clf_dict = defaultdict(list)\n",
    "    results = {}\n",
    "    csv_files = results_path.glob(\"**/*.csv\")\n",
    "    for csv in csv_files:\n",
    "        current_csv = str(csv).split(\"/\")\n",
    "        current_lemma = current_csv[1]\n",
    "        current_sense = current_csv[3].split(\"~\")[0]\n",
    "        lemma_pickle = pd.read_pickle(\"data/lemma_senses_\" + current_lemma + \".pickle\")\n",
    "        sst = lemma_pickle[lemma_pickle[\"id\"] == current_sense].iloc[0][\"daterange.start\"]\n",
    "            \n",
    "        try:\n",
    "            df = pd.read_csv(csv)\n",
    "            \n",
    "            df = df[[\"label\",\"year\",\"quotation_id\",\"bert_ts_\" + weighting + \"_centroid_sense_vector_blert_base_-1,-2,-3,-4_mean\",\n",
    "                     \"bert_ts_\" + weighting + \"_centroid_sense_vector_bert_base_-1,-2,-3,-4_mean\",\"bert_ts_\" + weighting + \"_centroid_sense_vector_bert_1850_-1,-2,-3,-4_mean\"]]\n",
    "            df = df.rename(columns={\"bert_ts_\" + weighting + \"_centroid_sense_vector_bert_base_-1,-2,-3,-4_mean\": \"bert_base_sense_centroid_\" + weighting,\n",
    "                                    \"bert_ts_\" + weighting + \"_centroid_sense_vector_bert_1850_-1,-2,-3,-4_mean\": \"bert_1850_sense_centroid_\" + weighting,\n",
    "                                    \"bert_ts_\" + weighting + \"_centroid_sense_vector_blert_base_-1,-2,-3,-4_mean\": \"bert_1900_sense_centroid_\" + weighting})\n",
    "            df = df[df[\"year\"].between(timestart, timestart+(year_window*2))]\n",
    "\n",
    "        except Exception as e:\n",
    "            continue\n",
    "        \n",
    "        if current_lemma in lemmas:\n",
    "            for col in df.columns:\n",
    "                clf_dict[col].extend(df[col])\n",
    "\n",
    "    for colname, classifications in clf_dict.items():\n",
    "        if colname not in ('label','year','quotation_id') and colname.startswith(\"bert_\"):\n",
    "            p,r = [round(x,3) for x in precision_recall_fscore_support(clf_dict['label'],classifications,average='binary',pos_label=1)[:2] if x] \n",
    "            f1 = round((2*(p*r))/(p+r),3)\n",
    "            if metric == \"recall\":\n",
    "                results[colname] = round(r,3)\n",
    "            if metric == \"precision\":\n",
    "                results[colname] = round(p,3)\n",
    "            if metric == \"fscore\":\n",
    "                results[colname] = round(f1,3)\n",
    "\n",
    "    return results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "setting = \"abstract\"\n",
    "weighting = \"weighted\"\n",
    "metric = \"fscore\"\n",
    "\n",
    "lemmas = dLemmas[setting]\n",
    "\n",
    "time_experiment = 1920 # Folder from which we select results: \"1850\" or \"1920\"\n",
    "test_daterange_start = 1760 # Quotations starting from\n",
    "year_window = 50\n",
    "bert_base = []\n",
    "bert_1850 = []\n",
    "bert_1900 = []\n",
    "time_mean = []\n",
    "for date_start in range(test_daterange_start, time_experiment - year_window,10):\n",
    "    results = find_optimal_daterange_ts(Path('results_ts_' + str(time_experiment)), lemmas, date_start, year_window, metric, weighting)\n",
    "    bert_base.append(results[\"bert_base_sense_centroid_\" + weighting])\n",
    "    bert_1850.append(results[\"bert_1850_sense_centroid_\" + weighting])\n",
    "    bert_1900.append(results[\"bert_1900_sense_centroid_\" + weighting])\n",
    "    time_mean.append(date_start + year_window)\n",
    "    \n",
    "plt.plot(time_mean, bert_base)\n",
    "plt.plot(time_mean, bert_1850)\n",
    "plt.plot(time_mean, bert_1900)\n",
    "\n",
    "plt.legend([\"BERTbase\", \"BERT1850\", \"BERT1900\"], loc='lower right')\n",
    "\n",
    "plt.savefig(\"figures/plot_LM_\" + setting + \"_class1\" + metric + \"_50_ts_\" + weighting + \".png\", dpi=300, bbox_inches=\"tight\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Inspect differences between time-sensitive and standard\n",
    "\n",
    "What do we gain by using the weighted method? And what do we lose with it? Compare quotations correclty labeled by the time-sensitive method and incorrectly labeled by the time-insensitive method, or viceversa."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "def time_sensitive_exploration(results_path, lemmas, observed_LM, ts_method, improvement):\n",
    "    from sklearn.metrics import precision_recall_fscore_support\n",
    "    clf_dict = defaultdict(list)\n",
    "    results = {}\n",
    "    csv_files = results_path.glob(\"**/*.csv\")\n",
    "    years = []\n",
    "    years_total = []\n",
    "    for csv in csv_files:\n",
    "        current_csv = str(csv).split(\"/\")\n",
    "        current_lemma = current_csv[1]\n",
    "        current_sense = current_csv[3].split(\"~\")[0]\n",
    "        lemma_pickle = pd.read_pickle(\"data/lemma_senses_\" + current_lemma + \".pickle\")\n",
    "        \n",
    "        if current_lemma in lemmas:\n",
    "            try:\n",
    "                df = pd.read_csv(csv)\n",
    "\n",
    "                df = df[[\"label\",\"year\",\"quotation_id\",\"bert_centroid_sense_vector_\" + observed_LM + \"_-1,-2,-3,-4_mean\",\n",
    "                         \"bert_ts_\" + ts_method + \"_centroid_sense_vector_\" + observed_LM + \"_-1,-2,-3,-4_mean\"]]\n",
    "\n",
    "                df = df.rename(columns={\"bert_centroid_sense_vector_\" + observed_LM + \"_-1,-2,-3,-4_mean\": observed_LM + \"_standard\",\n",
    "                                        \"bert_ts_\" + ts_method + \"_centroid_sense_vector_\" + observed_LM + \"_-1,-2,-3,-4_mean\": observed_LM + \"_\" + ts_method})\n",
    "\n",
    "            except Exception as e:\n",
    "                continue\n",
    "\n",
    "            filtered_df = pd.DataFrame()\n",
    "            if improvement == \"weighted\":\n",
    "                filtered_df = df[(df[\"label\"] == 1) & (df[observed_LM + \"_standard\"] == 0) & (df[observed_LM + \"_\" + ts_method] == 1)]\n",
    "            elif improvement == \"standard\":\n",
    "                filtered_df = df[(df[\"label\"] == 1) & (df[observed_LM + \"_standard\"] == 1) & (df[observed_LM + \"_\" + ts_method] == 0)]\n",
    "            \n",
    "            years += list(filtered_df[\"year\"].values)\n",
    "            years_total += list(df[df[\"label\"] == 1][\"year\"].values)\n",
    "    \n",
    "    batch1 = [y for y in years if y >= 1770 and y <= 1800]\n",
    "    batch2 = [y for y in years if y > 1800 and y <= 1830]\n",
    "    batch3 = [y for y in years if y > 1830 and y <= 1860]\n",
    "    batch4 = [y for y in years if y > 1860 and y <= 1890]\n",
    "    batch5 = [y for y in years if y > 1890 and y <= 1920]\n",
    "    \n",
    "    batch1_total = [y for y in years_total if y >= 1770 and y <= 1800]\n",
    "    batch2_total = [y for y in years_total if y > 1800 and y <= 1830]\n",
    "    batch3_total = [y for y in years_total if y > 1830 and y <= 1860]\n",
    "    batch4_total = [y for y in years_total if y > 1860 and y <= 1890]\n",
    "    batch5_total = [y for y in years_total if y > 1890 and y <= 1920]\n",
    "    \n",
    "    return len(batch1)/len(batch1_total), len(batch2)/len(batch2_total), len(batch3)/len(batch3_total), len(batch4)/len(batch4_total), len(batch5)/len(batch5_total)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Explore improvements by using: weighted\n",
      "==========================================\n",
      "Setting: abstract\n",
      "EXP: 1760-1920\n",
      "LM: bert_1850\n",
      "\tBatch 1750-1800: 0.05357142857142857\n",
      "\tBatch 1800-1850: 0.13725490196078433\n",
      "\tBatch 1850-1900: 0.07368421052631578\n",
      "\tBatch 1900-1950: 0.15306122448979592\n",
      "\tBatch 1950-2000: 0.125\n",
      "\tAll batches: 0.10851435310966492\n",
      "\n",
      "Setting: abstract\n",
      "EXP: 1760-1920\n",
      "LM: blert_base\n",
      "\tBatch 1750-1800: 0.05357142857142857\n",
      "\tBatch 1800-1850: 0.058823529411764705\n",
      "\tBatch 1850-1900: 0.09473684210526316\n",
      "\tBatch 1900-1950: 0.14285714285714285\n",
      "\tBatch 1950-2000: 0.140625\n",
      "\tAll batches: 0.09812278858911985\n",
      "\n",
      "Setting: abstract\n",
      "EXP: 1760-1920\n",
      "LM: bert_base\n",
      "\tBatch 1750-1800: 0.10714285714285714\n",
      "\tBatch 1800-1850: 0.1568627450980392\n",
      "\tBatch 1850-1900: 0.1368421052631579\n",
      "\tBatch 1900-1950: 0.15306122448979592\n",
      "\tBatch 1950-2000: 0.125\n",
      "\tAll batches: 0.13578178639877003\n",
      "\n",
      "Explore improvements by using: standard\n",
      "==========================================\n",
      "Setting: abstract\n",
      "EXP: 1760-1920\n",
      "LM: bert_1850\n",
      "\tBatch 1750-1800: 0.08928571428571429\n",
      "\tBatch 1800-1850: 0.09803921568627451\n",
      "\tBatch 1850-1900: 0.10526315789473684\n",
      "\tBatch 1900-1950: 0.08163265306122448\n",
      "\tBatch 1950-2000: 0.09375\n",
      "\tAll batches: 0.09359414818559002\n",
      "\n",
      "Setting: abstract\n",
      "EXP: 1760-1920\n",
      "LM: blert_base\n",
      "\tBatch 1750-1800: 0.10714285714285714\n",
      "\tBatch 1800-1850: 0.058823529411764705\n",
      "\tBatch 1850-1900: 0.06315789473684211\n",
      "\tBatch 1900-1950: 0.061224489795918366\n",
      "\tBatch 1950-2000: 0.0625\n",
      "\tAll batches: 0.07056975421747647\n",
      "\n",
      "Setting: abstract\n",
      "EXP: 1760-1920\n",
      "LM: bert_base\n",
      "\tBatch 1750-1800: 0.05357142857142857\n",
      "\tBatch 1800-1850: 0.058823529411764705\n",
      "\tBatch 1850-1900: 0.05263157894736842\n",
      "\tBatch 1900-1950: 0.07142857142857142\n",
      "\tBatch 1950-2000: 0.09375\n",
      "\tAll batches: 0.06604102167182663\n",
      "\n"
     ]
    }
   ],
   "source": [
    "setting = \"abstract\"\n",
    "metric = \"fscore\"\n",
    "improvement = \"weighted\" # \"weighted\" to see improvement of weighted over standard\n",
    "                         # \"standard\" to see improvement of standard over weighted\n",
    "\n",
    "lemmas = dLemmas[setting]\n",
    "\n",
    "lms = [\"bert_1850\", \"blert_base\", \"bert_base\"]\n",
    "\n",
    "for improvement in [\"weighted\", \"standard\"]:\n",
    "    print(\"Explore improvements by using:\", improvement)\n",
    "    print(\"==========================================\")\n",
    "    time_experiment = 1920 # Folder from which we select results\n",
    "    # observed_LM = \"bert_1850\"\n",
    "    for observed_LM in lms:\n",
    "        print(\"Setting:\", setting)\n",
    "        print(\"EXP: 1760-\" + str(time_experiment))\n",
    "        print(\"LM:\", observed_LM)\n",
    "        ts_method = \"weighted\"\n",
    "        batches = time_sensitive_exploration(Path('results_ts_' + str(time_experiment)), lemmas, observed_LM, ts_method, improvement)\n",
    "        print(\"\\tBatch 1750-1800:\", batches[0])\n",
    "        print(\"\\tBatch 1800-1850:\", batches[1])\n",
    "        print(\"\\tBatch 1850-1900:\", batches[2])\n",
    "        print(\"\\tBatch 1900-1950:\", batches[3])\n",
    "        print(\"\\tBatch 1950-2000:\", batches[4])\n",
    "        print(\"\\tAll batches:\", statistics.mean(batches))\n",
    "        print()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Data exploration\n",
    "\n",
    "Code to generate table 1 describing the data:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "### lemma: anger ###\n",
      "[LOG] # senses before filtering by date = 442\n",
      "[LOG] # senses after filtering by date = 198\n",
      "[LOG] # of seed senses 11 \n",
      "[LOG] # of synonyms 177 \n",
      "[LOG] # of branch senses 0\n",
      "[LOG] # of seeds selected 3 \n",
      "[LOG] # of synonyms selected 120 \n",
      "[LOG] # of branches selected 0\n",
      "[LOG] 4001 quotations selected\n",
      "[LOG] train = 3642 val = 126 test = 233 quotations\n",
      "### lemma: apple ###\n",
      "[LOG] # senses before filtering by date = 306\n",
      "[LOG] # senses after filtering by date = 170\n",
      "[LOG] # of seed senses 10 \n",
      "[LOG] # of synonyms 150 \n",
      "[LOG] # of branch senses 0\n",
      "[LOG] # of seeds selected 10 \n",
      "[LOG] # of synonyms selected 150 \n",
      "[LOG] # of branches selected 0\n",
      "[LOG] 6388 quotations selected\n",
      "[LOG] train = 5803 val = 209 test = 376 quotations\n",
      "### lemma: art ###\n",
      "[LOG] # senses before filtering by date = 266\n",
      "[LOG] # senses after filtering by date = 175\n",
      "[LOG] # of seed senses 19 \n",
      "[LOG] # of synonyms 135 \n",
      "[LOG] # of branch senses 0\n",
      "[LOG] # of seeds selected 17 \n",
      "[LOG] # of synonyms selected 121 \n",
      "[LOG] # of branches selected 0\n",
      "[LOG] 4655 quotations selected\n",
      "[LOG] train = 4230 val = 184 test = 241 quotations\n",
      "### lemma: democracy ###\n",
      "[LOG] # senses before filtering by date = 94\n",
      "[LOG] # senses after filtering by date = 64\n",
      "[LOG] # of seed senses 6 \n",
      "[LOG] # of synonyms 54 \n",
      "[LOG] # of branch senses 0\n",
      "[LOG] # of seeds selected 6 \n",
      "[LOG] # of synonyms selected 54 \n",
      "[LOG] # of branches selected 0\n",
      "[LOG] 2363 quotations selected\n",
      "[LOG] train = 2131 val = 72 test = 160 quotations\n",
      "### lemma: happiness ###\n",
      "[LOG] # senses before filtering by date = 126\n",
      "[LOG] # senses after filtering by date = 80\n",
      "[LOG] # of seed senses 5 \n",
      "[LOG] # of synonyms 70 \n",
      "[LOG] # of branch senses 0\n",
      "[LOG] # of seeds selected 5 \n",
      "[LOG] # of synonyms selected 70 \n",
      "[LOG] # of branches selected 0\n",
      "[LOG] 1218 quotations selected\n",
      "[LOG] train = 1115 val = 48 test = 55 quotations\n",
      "### lemma: labour ###\n",
      "[LOG] # senses before filtering by date = 210\n",
      "[LOG] # senses after filtering by date = 130\n",
      "[LOG] # of seed senses 13 \n",
      "[LOG] # of synonyms 105 \n",
      "[LOG] # of branch senses 0\n",
      "[LOG] # of seeds selected 13 \n",
      "[LOG] # of synonyms selected 105 \n",
      "[LOG] # of branches selected 0\n",
      "[LOG] 3189 quotations selected\n",
      "[LOG] train = 2891 val = 100 test = 198 quotations\n",
      "### lemma: machine ###\n",
      "[LOG] # senses before filtering by date = 517\n",
      "[LOG] # senses after filtering by date = 266\n",
      "[LOG] # of seed senses 17 \n",
      "[LOG] # of synonyms 234 \n",
      "[LOG] # of branch senses 0\n",
      "[LOG] # of seeds selected 17 \n",
      "[LOG] # of synonyms selected 234 \n",
      "[LOG] # of branches selected 0\n",
      "[LOG] 10081 quotations selected\n",
      "[LOG] train = 9076 val = 377 test = 628 quotations\n",
      "### lemma: man ###\n",
      "[LOG] # senses before filtering by date = 1662\n",
      "[LOG] # senses after filtering by date = 934\n",
      "[LOG] # of seed senses 67 \n",
      "[LOG] # of synonyms 814 \n",
      "[LOG] # of branch senses 0\n",
      "[LOG] # of seeds selected 55 \n",
      "[LOG] # of synonyms selected 562 \n",
      "[LOG] # of branches selected 0\n",
      "[LOG] 20897 quotations selected\n",
      "[LOG] train = 18978 val = 715 test = 1204 quotations\n",
      "### lemma: nation ###\n",
      "[LOG] # senses before filtering by date = 266\n",
      "[LOG] # senses after filtering by date = 163\n",
      "[LOG] # of seed senses 14 \n",
      "[LOG] # of synonyms 137 \n",
      "[LOG] # of branch senses 0\n",
      "[LOG] # of seeds selected 14 \n",
      "[LOG] # of synonyms selected 137 \n",
      "[LOG] # of branches selected 0\n",
      "[LOG] 7245 quotations selected\n",
      "[LOG] train = 6576 val = 233 test = 436 quotations\n",
      "### lemma: power ###\n",
      "[LOG] # senses before filtering by date = 717\n",
      "[LOG] # senses after filtering by date = 447\n",
      "[LOG] # of seed senses 46 \n",
      "[LOG] # of synonyms 356 \n",
      "[LOG] # of branch senses 0\n",
      "[LOG] # of seeds selected 30 \n",
      "[LOG] # of synonyms selected 257 \n",
      "[LOG] # of branches selected 0\n",
      "[LOG] 10855 quotations selected\n",
      "[LOG] train = 9825 val = 397 test = 633 quotations\n",
      "### lemma: slave ###\n",
      "[LOG] # senses before filtering by date = 637\n",
      "[LOG] # senses after filtering by date = 300\n",
      "[LOG] # of seed senses 9 \n",
      "[LOG] # of synonyms 280 \n",
      "[LOG] # of branch senses 0\n",
      "[LOG] # of seeds selected 8 \n",
      "[LOG] # of synonyms selected 273 \n",
      "[LOG] # of branches selected 0\n",
      "[LOG] 7730 quotations selected\n",
      "[LOG] train = 6990 val = 273 test = 467 quotations\n",
      "### lemma: woman ###\n",
      "[LOG] # senses before filtering by date = 597\n",
      "[LOG] # senses after filtering by date = 362\n",
      "[LOG] # of seed senses 15 \n",
      "[LOG] # of synonyms 329 \n",
      "[LOG] # of branch senses 0\n",
      "[LOG] # of seeds selected 15 \n",
      "[LOG] # of synonyms selected 329 \n",
      "[LOG] # of branches selected 0\n",
      "[LOG] 7532 quotations selected\n",
      "[LOG] train = 6806 val = 267 test = 459 quotations\n"
     ]
    }
   ],
   "source": [
    "import pandas as pd\n",
    "from utils.classificaton_utils import binarize\n",
    "\n",
    "words = [['anger',\"NN\"],[\"apple\",\"NN\"],[\"art\",\"NN\"],[\"democracy\",\"NN\"],\n",
    "         [\"happiness\",\"NN\"],[\"labour\",\"NN\"],[\"machine\",\"NN\"],[\"man\",\"NN\"],\n",
    "         [\"nation\",\"NN\"],[\"power\",\"NN\"],[\"slave\",\"NN\"],['woman','NN']]\n",
    "\n",
    "experiment = {\"start\": 1760, \"end\": 1850, \"filter_val\": True, \"filter_test\": True}\n",
    "\n",
    "dr = dict()\n",
    "for lemma, pos in words:\n",
    "    print(\"### lemma: {} ###\".format(lemma))\n",
    "    quotations_path = f\"./data/sfrel_quotations_{lemma}_{pos}.pickle\"\n",
    "    lemma_senses = pd.read_pickle(f'./data/lemma_senses_{lemma}_{pos}.pickle')\n",
    "\n",
    "    senses = set(lemma_senses[lemma_senses.word_id.str.startswith(f'{lemma}_{pos.lower()}')][\"id\"])\n",
    "    relations = ['seed','synonym']\n",
    "    eval_mode = \"lemma_etal\"\n",
    "\n",
    "    df_train, df_val, df_test = binarize(lemma=lemma,\n",
    "                                    pos=pos,\n",
    "                                    senses=senses, \n",
    "                                    start=experiment[\"start\"],\n",
    "                                    end=experiment[\"end\"],\n",
    "                                    relations=relations,\n",
    "                                    eval_mode=eval_mode,\n",
    "                                    filter_val_by_year=experiment[\"filter_val\"],\n",
    "                                    filter_test_by_year=experiment[\"filter_test\"],\n",
    "                                    strict_filter=True)\n",
    "\n",
    "    df_all = pd.concat([df_train, df_val, df_test])\n",
    "    \n",
    "    unique_seed_senses = len(df_all[df_all[\"provenance_type\"] == \"seed\"][\"sense_id\"].unique())\n",
    "    unique_syn_senses = round(len(df_all[df_all[\"provenance_type\"] == \"synonym\"][\"sense_id\"].unique())/unique_seed_senses)\n",
    "    unique_other_senses = round(len(df_all[~df_all[\"provenance_type\"].isin([\"synonym\", \"seed\"])][\"sense_id\"].unique())/unique_seed_senses)\n",
    "    df_all_posq = df_all[df_all[\"label\"] == \"1\"]\n",
    "    df_all_negq = df_all[df_all[\"label\"] == \"0\"]\n",
    "    quotations_p = round(len(df_all_posq.quotation_id.unique())/unique_seed_senses)\n",
    "    quotations_n = round(len(df_all_negq.quotation_id.unique())/unique_seed_senses)\n",
    "    quotations = str(quotations_p) + \"/\" + str(quotations_n)\n",
    "    derived_senses = str(unique_syn_senses) + \"/\" + str(unique_other_senses)\n",
    "     \n",
    "    dr[lemma] = [unique_seed_senses, derived_senses, quotations]\n",
    "    \n",
    "data_description = pd.DataFrame.from_dict(dr, orient='index', columns=['Seeds','ExpSenses', 'Quotations'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\\begin{tabular}{lrll}\n",
      "\\toprule\n",
      "{} &  Seeds & ExpSenses & Quotations \\\\\n",
      "\\midrule\n",
      "anger     &      6 &    17/121 &    103/564 \\\\\n",
      "apple     &     19 &      6/61 &     36/300 \\\\\n",
      "art       &     19 &      5/47 &     33/212 \\\\\n",
      "democracy &      7 &      6/61 &     37/301 \\\\\n",
      "happiness &      5 &      9/46 &     57/186 \\\\\n",
      "labour    &     18 &      4/30 &     29/148 \\\\\n",
      "machine   &     25 &      8/83 &     42/361 \\\\\n",
      "man       &     48 &      9/78 &     55/380 \\\\\n",
      "nation    &     15 &      8/85 &     53/430 \\\\\n",
      "power     &     39 &      5/49 &     34/244 \\\\\n",
      "slave     &     10 &    20/158 &    103/670 \\\\\n",
      "woman     &     17 &     10/81 &     64/379 \\\\\n",
      "\\bottomrule\n",
      "\\end{tabular}\n",
      "\n"
     ]
    }
   ],
   "source": [
    "print(data_description.to_latex())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python (py37_tsd)",
   "language": "python",
   "name": "py37_tsd"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.10"
  },
  "metadata": {
   "interpreter": {
    "hash": "5d002d29fc59799ec58342307c50be960f7297ba8a27d201097020f600ad239f"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}