4_eval_models_on_hatecheck.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# EVALUATING MODELS ON HATECHECK\n",
    "BERT models and commercial models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Initialise relevant packages\n",
    "\n",
    "# Basics\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "\n",
    "# Evaluation\n",
    "from sklearn.metrics import accuracy_score, precision_recall_fscore_support, classification_report"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load Test Suite"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# load test suite\n",
    "hatecheck_df = pd.read_csv('./Data/Test Suite/hatecheck_final_ACL.csv', index_col=0)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load and Merge Model Results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>functionality</th>\n",
       "      <th>case_id</th>\n",
       "      <th>test_case</th>\n",
       "      <th>label_gold</th>\n",
       "      <th>target_ident</th>\n",
       "      <th>direction</th>\n",
       "      <th>focus_words</th>\n",
       "      <th>focus_lemma</th>\n",
       "      <th>ref_case_id</th>\n",
       "      <th>ref_templ_id</th>\n",
       "      <th>templ_id</th>\n",
       "      <th>case_templ</th>\n",
       "      <th>pred_BERT_davidson2017_weighted</th>\n",
       "      <th>pred_BERT_founta2018_weighted</th>\n",
       "      <th>pred_perspective</th>\n",
       "      <th>pred_siftninja</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>derog_neg_emote_h</td>\n",
       "      <td>1</td>\n",
       "      <td>I hate women.</td>\n",
       "      <td>hateful</td>\n",
       "      <td>women</td>\n",
       "      <td>general</td>\n",
       "      <td>hate</td>\n",
       "      <td>hate</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1</td>\n",
       "      <td>I hate [IDENTITY_P].</td>\n",
       "      <td>hateful</td>\n",
       "      <td>hateful</td>\n",
       "      <td>hateful</td>\n",
       "      <td>non-hateful</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>derog_neg_emote_h</td>\n",
       "      <td>2</td>\n",
       "      <td>I hate trans people.</td>\n",
       "      <td>hateful</td>\n",
       "      <td>trans people</td>\n",
       "      <td>general</td>\n",
       "      <td>hate</td>\n",
       "      <td>hate</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1</td>\n",
       "      <td>I hate [IDENTITY_P].</td>\n",
       "      <td>hateful</td>\n",
       "      <td>hateful</td>\n",
       "      <td>hateful</td>\n",
       "      <td>hateful</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>derog_neg_emote_h</td>\n",
       "      <td>3</td>\n",
       "      <td>I hate gay people.</td>\n",
       "      <td>hateful</td>\n",
       "      <td>gay people</td>\n",
       "      <td>general</td>\n",
       "      <td>hate</td>\n",
       "      <td>hate</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1</td>\n",
       "      <td>I hate [IDENTITY_P].</td>\n",
       "      <td>hateful</td>\n",
       "      <td>hateful</td>\n",
       "      <td>hateful</td>\n",
       "      <td>hateful</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>derog_neg_emote_h</td>\n",
       "      <td>4</td>\n",
       "      <td>I hate black people.</td>\n",
       "      <td>hateful</td>\n",
       "      <td>black people</td>\n",
       "      <td>general</td>\n",
       "      <td>hate</td>\n",
       "      <td>hate</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1</td>\n",
       "      <td>I hate [IDENTITY_P].</td>\n",
       "      <td>hateful</td>\n",
       "      <td>hateful</td>\n",
       "      <td>hateful</td>\n",
       "      <td>hateful</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>derog_neg_emote_h</td>\n",
       "      <td>5</td>\n",
       "      <td>I hate disabled people.</td>\n",
       "      <td>hateful</td>\n",
       "      <td>disabled people</td>\n",
       "      <td>general</td>\n",
       "      <td>hate</td>\n",
       "      <td>hate</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1</td>\n",
       "      <td>I hate [IDENTITY_P].</td>\n",
       "      <td>hateful</td>\n",
       "      <td>hateful</td>\n",
       "      <td>hateful</td>\n",
       "      <td>non-hateful</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3723</th>\n",
       "      <td>spell_leet_h</td>\n",
       "      <td>3897</td>\n",
       "      <td>Gay people are all complete 4ssholes.</td>\n",
       "      <td>hateful</td>\n",
       "      <td>gay people</td>\n",
       "      <td>general</td>\n",
       "      <td>4ssholes</td>\n",
       "      <td>asshole</td>\n",
       "      <td>1232.0</td>\n",
       "      <td>256.0</td>\n",
       "      <td>866</td>\n",
       "      <td>[IDENTITY_P] are all complete 4ssholes.</td>\n",
       "      <td>hateful</td>\n",
       "      <td>hateful</td>\n",
       "      <td>hateful</td>\n",
       "      <td>non-hateful</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3724</th>\n",
       "      <td>spell_leet_h</td>\n",
       "      <td>3898</td>\n",
       "      <td>Black people are all complete 4ssholes.</td>\n",
       "      <td>hateful</td>\n",
       "      <td>black people</td>\n",
       "      <td>general</td>\n",
       "      <td>4ssholes</td>\n",
       "      <td>asshole</td>\n",
       "      <td>1233.0</td>\n",
       "      <td>256.0</td>\n",
       "      <td>866</td>\n",
       "      <td>[IDENTITY_P] are all complete 4ssholes.</td>\n",
       "      <td>hateful</td>\n",
       "      <td>hateful</td>\n",
       "      <td>hateful</td>\n",
       "      <td>non-hateful</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3725</th>\n",
       "      <td>spell_leet_h</td>\n",
       "      <td>3899</td>\n",
       "      <td>Disabled people are all complete 4ssholes.</td>\n",
       "      <td>hateful</td>\n",
       "      <td>disabled people</td>\n",
       "      <td>general</td>\n",
       "      <td>4ssholes</td>\n",
       "      <td>asshole</td>\n",
       "      <td>1234.0</td>\n",
       "      <td>256.0</td>\n",
       "      <td>866</td>\n",
       "      <td>[IDENTITY_P] are all complete 4ssholes.</td>\n",
       "      <td>hateful</td>\n",
       "      <td>non-hateful</td>\n",
       "      <td>hateful</td>\n",
       "      <td>non-hateful</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3726</th>\n",
       "      <td>spell_leet_h</td>\n",
       "      <td>3900</td>\n",
       "      <td>Muslims are all complete 4ssholes.</td>\n",
       "      <td>hateful</td>\n",
       "      <td>Muslims</td>\n",
       "      <td>general</td>\n",
       "      <td>4ssholes</td>\n",
       "      <td>asshole</td>\n",
       "      <td>1235.0</td>\n",
       "      <td>256.0</td>\n",
       "      <td>866</td>\n",
       "      <td>[IDENTITY_P] are all complete 4ssholes.</td>\n",
       "      <td>hateful</td>\n",
       "      <td>hateful</td>\n",
       "      <td>hateful</td>\n",
       "      <td>non-hateful</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3727</th>\n",
       "      <td>spell_leet_h</td>\n",
       "      <td>3901</td>\n",
       "      <td>Immigrants are all complete 4ssholes.</td>\n",
       "      <td>hateful</td>\n",
       "      <td>immigrants</td>\n",
       "      <td>general</td>\n",
       "      <td>4ssholes</td>\n",
       "      <td>asshole</td>\n",
       "      <td>1236.0</td>\n",
       "      <td>256.0</td>\n",
       "      <td>866</td>\n",
       "      <td>[IDENTITY_P] are all complete 4ssholes.</td>\n",
       "      <td>hateful</td>\n",
       "      <td>non-hateful</td>\n",
       "      <td>hateful</td>\n",
       "      <td>non-hateful</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>3728 rows × 16 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "          functionality  case_id                                    test_case  \\\n",
       "0     derog_neg_emote_h        1                               I hate women.    \n",
       "1     derog_neg_emote_h        2                        I hate trans people.    \n",
       "2     derog_neg_emote_h        3                          I hate gay people.    \n",
       "3     derog_neg_emote_h        4                        I hate black people.    \n",
       "4     derog_neg_emote_h        5                     I hate disabled people.    \n",
       "...                 ...      ...                                          ...   \n",
       "3723       spell_leet_h     3897       Gay people are all complete 4ssholes.    \n",
       "3724       spell_leet_h     3898     Black people are all complete 4ssholes.    \n",
       "3725       spell_leet_h     3899  Disabled people are all complete 4ssholes.    \n",
       "3726       spell_leet_h     3900          Muslims are all complete 4ssholes.    \n",
       "3727       spell_leet_h     3901       Immigrants are all complete 4ssholes.    \n",
       "\n",
       "     label_gold     target_ident direction focus_words focus_lemma  \\\n",
       "0       hateful            women   general        hate        hate   \n",
       "1       hateful     trans people   general        hate        hate   \n",
       "2       hateful       gay people   general        hate        hate   \n",
       "3       hateful     black people   general        hate        hate   \n",
       "4       hateful  disabled people   general        hate        hate   \n",
       "...         ...              ...       ...         ...         ...   \n",
       "3723    hateful       gay people   general    4ssholes     asshole   \n",
       "3724    hateful     black people   general    4ssholes     asshole   \n",
       "3725    hateful  disabled people   general    4ssholes     asshole   \n",
       "3726    hateful          Muslims   general    4ssholes     asshole   \n",
       "3727    hateful       immigrants   general    4ssholes     asshole   \n",
       "\n",
       "      ref_case_id  ref_templ_id  templ_id  \\\n",
       "0             NaN           NaN         1   \n",
       "1             NaN           NaN         1   \n",
       "2             NaN           NaN         1   \n",
       "3             NaN           NaN         1   \n",
       "4             NaN           NaN         1   \n",
       "...           ...           ...       ...   \n",
       "3723       1232.0         256.0       866   \n",
       "3724       1233.0         256.0       866   \n",
       "3725       1234.0         256.0       866   \n",
       "3726       1235.0         256.0       866   \n",
       "3727       1236.0         256.0       866   \n",
       "\n",
       "                                   case_templ pred_BERT_davidson2017_weighted  \\\n",
       "0                        I hate [IDENTITY_P].                         hateful   \n",
       "1                        I hate [IDENTITY_P].                         hateful   \n",
       "2                        I hate [IDENTITY_P].                         hateful   \n",
       "3                        I hate [IDENTITY_P].                         hateful   \n",
       "4                        I hate [IDENTITY_P].                         hateful   \n",
       "...                                       ...                             ...   \n",
       "3723  [IDENTITY_P] are all complete 4ssholes.                         hateful   \n",
       "3724  [IDENTITY_P] are all complete 4ssholes.                         hateful   \n",
       "3725  [IDENTITY_P] are all complete 4ssholes.                         hateful   \n",
       "3726  [IDENTITY_P] are all complete 4ssholes.                         hateful   \n",
       "3727  [IDENTITY_P] are all complete 4ssholes.                         hateful   \n",
       "\n",
       "     pred_BERT_founta2018_weighted pred_perspective pred_siftninja  \n",
       "0                          hateful          hateful    non-hateful  \n",
       "1                          hateful          hateful        hateful  \n",
       "2                          hateful          hateful        hateful  \n",
       "3                          hateful          hateful        hateful  \n",
       "4                          hateful          hateful    non-hateful  \n",
       "...                            ...              ...            ...  \n",
       "3723                       hateful          hateful    non-hateful  \n",
       "3724                       hateful          hateful    non-hateful  \n",
       "3725                   non-hateful          hateful    non-hateful  \n",
       "3726                       hateful          hateful    non-hateful  \n",
       "3727                   non-hateful          hateful    non-hateful  \n",
       "\n",
       "[3728 rows x 16 columns]"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# load results\n",
    "results = {}\n",
    "results['BERT'] = pd.read_pickle('./Data/Test Suite/results_BERT_weighted_ACL.pkl')\n",
    "results['commercial'] = pd.read_pickle('./Data/Test Suite/results_commercial_models_ACL.pkl')\n",
    "\n",
    "\n",
    "# merge with hatecheck df\n",
    "for model in results:\n",
    "    hatecheck_df = hatecheck_df.merge(results[model], how = 'left', on = 'case_id')\n",
    "\n",
    "hatecheck_df"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Compute Accuracy by Functionality across Models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# set of models to evaluate\n",
    "models_eval = ['BERT_davidson2017_weighted','BERT_founta2018_weighted',\n",
    "               'perspective', 'siftninja']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# write data to dict\n",
    "func_accuracy_dict = {}\n",
    "\n",
    "for m in models_eval:\n",
    "    func_accuracy_dict[m] = []\n",
    "    for func in pd.unique(hatecheck_df.functionality):\n",
    "        n_cases = hatecheck_df[hatecheck_df.functionality==func].shape[0]\n",
    "        n_correct = hatecheck_df[(hatecheck_df.functionality==func)&(hatecheck_df['label_gold']==hatecheck_df['pred_{}'.format(m)])].shape[0]\n",
    "        func_accuracy_dict[m].append('{:.1%}'.format(n_correct/n_cases))\n",
    "    \n",
    "    # convert list to series\n",
    "    func_accuracy_dict[m] = pd.Series(func_accuracy_dict[m])\n",
    "    func_accuracy_dict[m].name = m"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# create df from dict\n",
    "func_accuracy_df = pd.Series(pd.unique(hatecheck_df.functionality))\n",
    "func_accuracy_df.name = 'functionality'\n",
    "\n",
    "for arc_data in func_accuracy_dict:\n",
    "    func_accuracy_df = pd.concat([func_accuracy_df, pd.Series(func_accuracy_dict[arc_data])], axis =1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>functionality</th>\n",
       "      <th>BERT_davidson2017_weighted</th>\n",
       "      <th>BERT_founta2018_weighted</th>\n",
       "      <th>perspective</th>\n",
       "      <th>siftninja</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>derog_neg_emote_h</td>\n",
       "      <td>88.6%</td>\n",
       "      <td>90.7%</td>\n",
       "      <td>98.6%</td>\n",
       "      <td>7.9%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>derog_neg_attrib_h</td>\n",
       "      <td>88.6%</td>\n",
       "      <td>84.3%</td>\n",
       "      <td>95.7%</td>\n",
       "      <td>0.0%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>derog_dehum_h</td>\n",
       "      <td>91.4%</td>\n",
       "      <td>80.7%</td>\n",
       "      <td>98.6%</td>\n",
       "      <td>10.0%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>derog_impl_h</td>\n",
       "      <td>71.4%</td>\n",
       "      <td>61.4%</td>\n",
       "      <td>85.0%</td>\n",
       "      <td>0.0%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>threat_dir_h</td>\n",
       "      <td>87.2%</td>\n",
       "      <td>75.9%</td>\n",
       "      <td>100.0%</td>\n",
       "      <td>4.5%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>threat_norm_h</td>\n",
       "      <td>91.4%</td>\n",
       "      <td>83.6%</td>\n",
       "      <td>100.0%</td>\n",
       "      <td>2.1%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>slur_h</td>\n",
       "      <td>60.4%</td>\n",
       "      <td>41.0%</td>\n",
       "      <td>66.0%</td>\n",
       "      <td>50.0%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>slur_homonym_nh</td>\n",
       "      <td>66.7%</td>\n",
       "      <td>70.0%</td>\n",
       "      <td>63.3%</td>\n",
       "      <td>33.3%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>slur_reclaimed_nh</td>\n",
       "      <td>39.5%</td>\n",
       "      <td>33.3%</td>\n",
       "      <td>28.4%</td>\n",
       "      <td>18.5%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>profanity_h</td>\n",
       "      <td>82.9%</td>\n",
       "      <td>72.9%</td>\n",
       "      <td>100.0%</td>\n",
       "      <td>1.4%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>profanity_nh</td>\n",
       "      <td>99.0%</td>\n",
       "      <td>100.0%</td>\n",
       "      <td>98.0%</td>\n",
       "      <td>100.0%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>ref_subs_clause_h</td>\n",
       "      <td>87.1%</td>\n",
       "      <td>80.7%</td>\n",
       "      <td>99.3%</td>\n",
       "      <td>0.7%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>ref_subs_sent_h</td>\n",
       "      <td>85.7%</td>\n",
       "      <td>70.7%</td>\n",
       "      <td>100.0%</td>\n",
       "      <td>0.0%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>negate_pos_h</td>\n",
       "      <td>85.0%</td>\n",
       "      <td>60.7%</td>\n",
       "      <td>96.4%</td>\n",
       "      <td>2.9%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>negate_neg_nh</td>\n",
       "      <td>12.8%</td>\n",
       "      <td>12.0%</td>\n",
       "      <td>3.8%</td>\n",
       "      <td>96.2%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15</th>\n",
       "      <td>phrase_question_h</td>\n",
       "      <td>80.7%</td>\n",
       "      <td>75.0%</td>\n",
       "      <td>99.3%</td>\n",
       "      <td>9.3%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16</th>\n",
       "      <td>phrase_opinion_h</td>\n",
       "      <td>85.7%</td>\n",
       "      <td>75.9%</td>\n",
       "      <td>98.5%</td>\n",
       "      <td>2.3%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17</th>\n",
       "      <td>ident_neutral_nh</td>\n",
       "      <td>20.6%</td>\n",
       "      <td>58.7%</td>\n",
       "      <td>84.1%</td>\n",
       "      <td>100.0%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18</th>\n",
       "      <td>ident_pos_nh</td>\n",
       "      <td>21.7%</td>\n",
       "      <td>52.9%</td>\n",
       "      <td>54.0%</td>\n",
       "      <td>100.0%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19</th>\n",
       "      <td>counter_quote_nh</td>\n",
       "      <td>26.6%</td>\n",
       "      <td>32.9%</td>\n",
       "      <td>15.6%</td>\n",
       "      <td>79.8%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20</th>\n",
       "      <td>counter_ref_nh</td>\n",
       "      <td>29.1%</td>\n",
       "      <td>29.8%</td>\n",
       "      <td>18.4%</td>\n",
       "      <td>79.4%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>21</th>\n",
       "      <td>target_obj_nh</td>\n",
       "      <td>87.7%</td>\n",
       "      <td>84.6%</td>\n",
       "      <td>95.4%</td>\n",
       "      <td>100.0%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>22</th>\n",
       "      <td>target_indiv_nh</td>\n",
       "      <td>27.7%</td>\n",
       "      <td>55.4%</td>\n",
       "      <td>84.6%</td>\n",
       "      <td>100.0%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>23</th>\n",
       "      <td>target_group_nh</td>\n",
       "      <td>35.5%</td>\n",
       "      <td>59.7%</td>\n",
       "      <td>62.9%</td>\n",
       "      <td>98.4%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>24</th>\n",
       "      <td>spell_char_swap_h</td>\n",
       "      <td>69.9%</td>\n",
       "      <td>58.6%</td>\n",
       "      <td>88.7%</td>\n",
       "      <td>11.3%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25</th>\n",
       "      <td>spell_char_del_h</td>\n",
       "      <td>59.3%</td>\n",
       "      <td>47.9%</td>\n",
       "      <td>74.3%</td>\n",
       "      <td>0.0%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>26</th>\n",
       "      <td>spell_space_del_h</td>\n",
       "      <td>68.1%</td>\n",
       "      <td>51.1%</td>\n",
       "      <td>80.1%</td>\n",
       "      <td>13.5%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>27</th>\n",
       "      <td>spell_space_add_h</td>\n",
       "      <td>43.9%</td>\n",
       "      <td>37.6%</td>\n",
       "      <td>74.0%</td>\n",
       "      <td>22.5%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>28</th>\n",
       "      <td>spell_leet_h</td>\n",
       "      <td>48.0%</td>\n",
       "      <td>43.9%</td>\n",
       "      <td>68.2%</td>\n",
       "      <td>16.2%</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         functionality BERT_davidson2017_weighted BERT_founta2018_weighted  \\\n",
       "0    derog_neg_emote_h                      88.6%                    90.7%   \n",
       "1   derog_neg_attrib_h                      88.6%                    84.3%   \n",
       "2        derog_dehum_h                      91.4%                    80.7%   \n",
       "3         derog_impl_h                      71.4%                    61.4%   \n",
       "4         threat_dir_h                      87.2%                    75.9%   \n",
       "5        threat_norm_h                      91.4%                    83.6%   \n",
       "6               slur_h                      60.4%                    41.0%   \n",
       "7      slur_homonym_nh                      66.7%                    70.0%   \n",
       "8    slur_reclaimed_nh                      39.5%                    33.3%   \n",
       "9          profanity_h                      82.9%                    72.9%   \n",
       "10        profanity_nh                      99.0%                   100.0%   \n",
       "11   ref_subs_clause_h                      87.1%                    80.7%   \n",
       "12     ref_subs_sent_h                      85.7%                    70.7%   \n",
       "13        negate_pos_h                      85.0%                    60.7%   \n",
       "14       negate_neg_nh                      12.8%                    12.0%   \n",
       "15   phrase_question_h                      80.7%                    75.0%   \n",
       "16    phrase_opinion_h                      85.7%                    75.9%   \n",
       "17    ident_neutral_nh                      20.6%                    58.7%   \n",
       "18        ident_pos_nh                      21.7%                    52.9%   \n",
       "19    counter_quote_nh                      26.6%                    32.9%   \n",
       "20      counter_ref_nh                      29.1%                    29.8%   \n",
       "21       target_obj_nh                      87.7%                    84.6%   \n",
       "22     target_indiv_nh                      27.7%                    55.4%   \n",
       "23     target_group_nh                      35.5%                    59.7%   \n",
       "24   spell_char_swap_h                      69.9%                    58.6%   \n",
       "25    spell_char_del_h                      59.3%                    47.9%   \n",
       "26   spell_space_del_h                      68.1%                    51.1%   \n",
       "27   spell_space_add_h                      43.9%                    37.6%   \n",
       "28        spell_leet_h                      48.0%                    43.9%   \n",
       "\n",
       "   perspective siftninja  \n",
       "0        98.6%      7.9%  \n",
       "1        95.7%      0.0%  \n",
       "2        98.6%     10.0%  \n",
       "3        85.0%      0.0%  \n",
       "4       100.0%      4.5%  \n",
       "5       100.0%      2.1%  \n",
       "6        66.0%     50.0%  \n",
       "7        63.3%     33.3%  \n",
       "8        28.4%     18.5%  \n",
       "9       100.0%      1.4%  \n",
       "10       98.0%    100.0%  \n",
       "11       99.3%      0.7%  \n",
       "12      100.0%      0.0%  \n",
       "13       96.4%      2.9%  \n",
       "14        3.8%     96.2%  \n",
       "15       99.3%      9.3%  \n",
       "16       98.5%      2.3%  \n",
       "17       84.1%    100.0%  \n",
       "18       54.0%    100.0%  \n",
       "19       15.6%     79.8%  \n",
       "20       18.4%     79.4%  \n",
       "21       95.4%    100.0%  \n",
       "22       84.6%    100.0%  \n",
       "23       62.9%     98.4%  \n",
       "24       88.7%     11.3%  \n",
       "25       74.3%      0.0%  \n",
       "26       80.1%     13.5%  \n",
       "27       74.0%     22.5%  \n",
       "28       68.2%     16.2%  "
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "func_accuracy_df"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Compute Classification Reports for Each Model "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "BERT_DAVIDSON2017_WEIGHTED\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      " non-hateful      0.401     0.360     0.379      1165\n",
      "     hateful      0.722     0.755     0.738      2563\n",
      "\n",
      "    accuracy                          0.632      3728\n",
      "   macro avg      0.561     0.558     0.559      3728\n",
      "weighted avg      0.621     0.632     0.626      3728\n",
      "\n",
      "\n",
      "BERT_FOUNTA2018_WEIGHTED\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      " non-hateful      0.390     0.485     0.432      1165\n",
      "     hateful      0.737     0.655     0.694      2563\n",
      "\n",
      "    accuracy                          0.602      3728\n",
      "   macro avg      0.563     0.570     0.563      3728\n",
      "weighted avg      0.628     0.602     0.612      3728\n",
      "\n",
      "\n",
      "PERSPECTIVE\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      " non-hateful      0.677     0.482     0.563      1165\n",
      "     hateful      0.792     0.895     0.841      2563\n",
      "\n",
      "    accuracy                          0.766      3728\n",
      "   macro avg      0.735     0.689     0.702      3728\n",
      "weighted avg      0.756     0.766     0.754      3728\n",
      "\n",
      "\n",
      "SIFTNINJA\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      " non-hateful      0.302     0.866     0.448      1165\n",
      "     hateful      0.596     0.090     0.156      2563\n",
      "\n",
      "    accuracy                          0.332      3728\n",
      "   macro avg      0.449     0.478     0.302      3728\n",
      "weighted avg      0.504     0.332     0.247      3728\n",
      "\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# print classification reports for each classifier\n",
    "\n",
    "for m in models_eval:\n",
    "        print(m.upper())\n",
    "        print(classification_report(hatecheck_df.label_gold.replace({'hateful': 1, 'non-hateful':0}),\n",
    "                                    hatecheck_df['pred_{}'.format(m)].replace({'hateful': 1, 'non-hateful':0}),\n",
    "                                    target_names = ['non-hateful','hateful'],\n",
    "                                    digits=3))\n",
    "        print()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ACCURACY \n",
      "\n",
      "BERT_davidson2017_weighted\n",
      "hateful: 75.5%:\n",
      "non-hateful: 36.0%:\n",
      "TOTAL: 63.2%\n",
      "\n",
      "BERT_founta2018_weighted\n",
      "hateful: 65.5%:\n",
      "non-hateful: 48.5%:\n",
      "TOTAL: 60.2%\n",
      "\n",
      "perspective\n",
      "hateful: 89.5%:\n",
      "non-hateful: 48.2%:\n",
      "TOTAL: 76.6%\n",
      "\n",
      "siftninja\n",
      "hateful: 9.0%:\n",
      "non-hateful: 86.6%:\n",
      "TOTAL: 33.2%\n",
      "\n"
     ]
    }
   ],
   "source": [
    "print('ACCURACY \\n')\n",
    "for m in models_eval:\n",
    "    print(m)\n",
    "    for lab in ['hateful', 'non-hateful']:\n",
    "        n_cases = hatecheck_df[hatecheck_df.label_gold==lab].shape[0]\n",
    "        n_correct = hatecheck_df[(hatecheck_df.label_gold==lab) & (hatecheck_df['label_gold']==hatecheck_df['pred_{}'.format(m)])].shape[0]\n",
    "        print('{}: {:.1%}:'.format(lab, n_correct/n_cases))\n",
    "    print('TOTAL: {:.1%}'.format(hatecheck_df[(hatecheck_df['label_gold']==hatecheck_df['pred_{}'.format(m)])].shape[0]/hatecheck_df.shape[0]))\n",
    "    print()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Performance for Reclaimed Slurs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ACCURACY\n",
      "BERT_DAVIDSON2017_WEIGHTED\n",
      "19\n",
      "nigga : 89.5%\n",
      "16\n",
      "fag : 0.0%\n",
      "16\n",
      "faggot : 0.0%\n",
      "15\n",
      "queer : 0.0%\n",
      "15\n",
      "bitch : 100.0%\n",
      "\n",
      "BERT_FOUNTA2018_WEIGHTED\n",
      "19\n",
      "nigga : 0.0%\n",
      "16\n",
      "fag : 6.2%\n",
      "16\n",
      "faggot : 6.2%\n",
      "15\n",
      "queer : 73.3%\n",
      "15\n",
      "bitch : 93.3%\n",
      "\n",
      "PERSPECTIVE\n",
      "19\n",
      "nigga : 0.0%\n",
      "16\n",
      "fag : 0.0%\n",
      "16\n",
      "faggot : 0.0%\n",
      "15\n",
      "queer : 80.0%\n",
      "15\n",
      "bitch : 73.3%\n",
      "\n",
      "SIFTNINJA\n",
      "19\n",
      "nigga : 0.0%\n",
      "16\n",
      "fag : 0.0%\n",
      "16\n",
      "faggot : 0.0%\n",
      "15\n",
      "queer : 0.0%\n",
      "15\n",
      "bitch : 100.0%\n",
      "\n"
     ]
    }
   ],
   "source": [
    "hatecheck_df[(hatecheck_df.functionality == 'slur_reclaimed_nh')].groupby('focus_lemma').case_id.count().keys()\n",
    "\n",
    "print('ACCURACY')\n",
    "for m in models_eval:\n",
    "    print(m.upper())\n",
    "    for slur in ['nigga', 'fag', 'faggot', 'queer', 'bitch']:\n",
    "        n_total = hatecheck_df[(hatecheck_df.functionality == 'slur_reclaimed_nh')&\n",
    "                                (hatecheck_df.focus_lemma==slur)].shape[0]\n",
    "        n_correct = hatecheck_df[(hatecheck_df.functionality == 'slur_reclaimed_nh')&\n",
    "                                  (hatecheck_df['pred_{}'.format(m)]==hatecheck_df.label_gold)&\n",
    "                                  (hatecheck_df.focus_lemma==slur)].shape[0]\n",
    "        print(n_total)\n",
    "        print(slur, ': {:.1%}'.format(n_correct/n_total))\n",
    "    print()\n",
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Performance Across Cases by Target Identity\n",
    "Only uses cases generated from templates"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "target_ident\n",
       "Muslims            421\n",
       "black people       421\n",
       "disabled people    421\n",
       "gay people         421\n",
       "immigrants         421\n",
       "trans people       421\n",
       "women              421\n",
       "Name: case_id, dtype: int64"
      ]
     },
     "execution_count": 50,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# create df with only template cases --> number of cases for each identity should be balanced\n",
    "templ_cases_df = hatecheck_df[hatecheck_df.case_templ.str.contains('IDENTITY')].copy()\n",
    "\n",
    "templ_cases_df.groupby(templ_cases_df.target_ident).case_id.count()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "# write data to dict\n",
    "ident_accuracy_dict = {}\n",
    "\n",
    "for m in models_eval:\n",
    "    ident_accuracy_dict[m] = []\n",
    "    for ident in pd.unique(templ_cases_df.target_ident):\n",
    "        n_cases = templ_cases_df[templ_cases_df.target_ident==ident].shape[0]\n",
    "        n_correct = templ_cases_df[(templ_cases_df.target_ident==ident)&(templ_cases_df['label_gold']==templ_cases_df['pred_{}'.format(m)])].shape[0]\n",
    "        ident_accuracy_dict[m].append('{:.1%}'.format(n_correct/n_cases))\n",
    "    ident_accuracy_dict[m] = pd.Series(ident_accuracy_dict[m])\n",
    "    ident_accuracy_dict[m].name = m"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "# create df from dict\n",
    "ident_accuracy_df = pd.Series(pd.unique(templ_cases_df.target_ident))\n",
    "ident_accuracy_df.name = 'target_ident'\n",
    "\n",
    "for arc_data in ident_accuracy_dict:\n",
    "    ident_accuracy_df = pd.concat([ident_accuracy_df, pd.Series(ident_accuracy_dict[arc_data])], axis =1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>target_ident</th>\n",
       "      <th>BERT_davidson2017_weighted</th>\n",
       "      <th>BERT_founta2018_weighted</th>\n",
       "      <th>perspective</th>\n",
       "      <th>siftninja</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>women</td>\n",
       "      <td>34.9%</td>\n",
       "      <td>52.3%</td>\n",
       "      <td>80.5%</td>\n",
       "      <td>23.0%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>trans people</td>\n",
       "      <td>69.1%</td>\n",
       "      <td>69.4%</td>\n",
       "      <td>80.8%</td>\n",
       "      <td>26.4%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>gay people</td>\n",
       "      <td>73.9%</td>\n",
       "      <td>74.3%</td>\n",
       "      <td>80.8%</td>\n",
       "      <td>25.9%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>black people</td>\n",
       "      <td>69.8%</td>\n",
       "      <td>72.2%</td>\n",
       "      <td>80.5%</td>\n",
       "      <td>26.6%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>disabled people</td>\n",
       "      <td>71.0%</td>\n",
       "      <td>37.1%</td>\n",
       "      <td>79.8%</td>\n",
       "      <td>23.0%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>Muslims</td>\n",
       "      <td>72.2%</td>\n",
       "      <td>73.6%</td>\n",
       "      <td>79.6%</td>\n",
       "      <td>27.6%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>immigrants</td>\n",
       "      <td>70.5%</td>\n",
       "      <td>58.9%</td>\n",
       "      <td>80.5%</td>\n",
       "      <td>25.9%</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      target_ident BERT_davidson2017_weighted BERT_founta2018_weighted  \\\n",
       "0            women                      34.9%                    52.3%   \n",
       "1     trans people                      69.1%                    69.4%   \n",
       "2       gay people                      73.9%                    74.3%   \n",
       "3     black people                      69.8%                    72.2%   \n",
       "4  disabled people                      71.0%                    37.1%   \n",
       "5          Muslims                      72.2%                    73.6%   \n",
       "6       immigrants                      70.5%                    58.9%   \n",
       "\n",
       "  perspective siftninja  \n",
       "0       80.5%     23.0%  \n",
       "1       80.8%     26.4%  \n",
       "2       80.8%     25.9%  \n",
       "3       80.5%     26.6%  \n",
       "4       79.8%     23.0%  \n",
       "5       79.6%     27.6%  \n",
       "6       80.5%     25.9%  "
      ]
     },
     "execution_count": 53,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ident_accuracy_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.8.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}