cicc_atis_bert.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "ddb87912-eef5-4e35-9ab7-b070caf1eb64",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "import math\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "from mapie.classification import MapieClassifier\n",
    "from mapie.metrics import classification_coverage_score, classification_mean_width_score\n",
    "import matplotlib.pyplot as plt\n",
    "import json\n",
    "\n",
    "from src.model import MockModel\n",
    "from src.utils import index_for_arange, write_results, cq_filter, cq_labels\n",
    "import src.plot as plot\n",
    "from src.eval import (\n",
    "    adjusted_classification_x_width_score,\n",
    "    size_stratified_classification_coverage_score,\n",
    "    cq_stats,\n",
    "    set_size_equals,\n",
    ")\n",
    "from src.heuristics import TopKSelector, HeuristicCutoffSelector, TopKCutoffSelector"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "8ab6a230-0e0d-422d-876b-0d85b4cb074a",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "26"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# load data\n",
    "dataset = 'atis'\n",
    "intent_classifier = 'bert'\n",
    "data_dir = f'data/{dataset}/{intent_classifier}/'\n",
    "\n",
    "df_train_preds = pd.read_csv(data_dir + \"train_predictions.csv\")\n",
    "df_test_preds = pd.read_csv(data_dir + \"test_predictions.csv\")\n",
    "df_calibration_preds = pd.read_csv(data_dir + \"calibration_predictions.csv\")\n",
    "\n",
    "labels = pd.read_csv(data_dir+'/labels.csv', header=0, names=['index','label'])\n",
    "labels = list(labels.label)\n",
    "N_INTENTS = len(labels)\n",
    "N_INTENTS"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "894d9cd0-64d3-425d-889a-dcd0055be9d7",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(3522, 1175, 1174)"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# check train-calibration-test split sizes\n",
    "df_train_preds.shape[0], df_calibration_preds.shape[0], df_test_preds.shape[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "84a844f0-da86-4e9f-94d6-e603194a6cd5",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "# apply some basic transformations\n",
    "df_train = df_train_preds.reset_index().rename(columns={'index': 'utterance'})\n",
    "df_test = df_test_preds.reset_index().rename(columns={'index': 'utterance'})\n",
    "df_calibration = df_calibration_preds.reset_index().rename(columns={'index': 'utterance'})\n",
    "\n",
    "df_train['utterance'] = df_train['utterance'].apply(lambda h: 'train_' + str(h))\n",
    "df_test['utterance'] = df_test['utterance'].apply(lambda h: 'test_' + str(h))\n",
    "df_calibration['utterance'] = df_calibration['utterance'].apply(lambda h: 'cal_' + str(h))\n",
    "\n",
    "df_train['utterance'] = df_train['utterance'].astype(str)\n",
    "df_test['utterance'] = df_test['utterance'].astype(str)\n",
    "df_calibration['utterance'] = df_calibration['utterance'].astype(str)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "a1dd2945-3fb7-4ca7-9325-397719061827",
   "metadata": {
    "tags": []
   },
   "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>utterance</th>\n",
       "      <th>label_0</th>\n",
       "      <th>label_1</th>\n",
       "      <th>label_2</th>\n",
       "      <th>label_3</th>\n",
       "      <th>label_4</th>\n",
       "      <th>label_5</th>\n",
       "      <th>label_6</th>\n",
       "      <th>label_7</th>\n",
       "      <th>label_8</th>\n",
       "      <th>...</th>\n",
       "      <th>label_16</th>\n",
       "      <th>label_17</th>\n",
       "      <th>label_18</th>\n",
       "      <th>label_19</th>\n",
       "      <th>label_20</th>\n",
       "      <th>label_21</th>\n",
       "      <th>label_22</th>\n",
       "      <th>label_23</th>\n",
       "      <th>label_24</th>\n",
       "      <th>label_25</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>test_0</td>\n",
       "      <td>0.005376</td>\n",
       "      <td>0.020964</td>\n",
       "      <td>0.008766</td>\n",
       "      <td>0.001087</td>\n",
       "      <td>0.010478</td>\n",
       "      <td>0.004582</td>\n",
       "      <td>0.003384</td>\n",
       "      <td>0.009285</td>\n",
       "      <td>0.770127</td>\n",
       "      <td>...</td>\n",
       "      <td>0.007280</td>\n",
       "      <td>0.007319</td>\n",
       "      <td>0.010551</td>\n",
       "      <td>0.006741</td>\n",
       "      <td>0.016093</td>\n",
       "      <td>0.029515</td>\n",
       "      <td>0.012482</td>\n",
       "      <td>0.012054</td>\n",
       "      <td>0.002456</td>\n",
       "      <td>0.008578</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>test_1</td>\n",
       "      <td>0.000051</td>\n",
       "      <td>0.000076</td>\n",
       "      <td>0.000036</td>\n",
       "      <td>0.000105</td>\n",
       "      <td>0.000040</td>\n",
       "      <td>0.000039</td>\n",
       "      <td>0.000062</td>\n",
       "      <td>0.000067</td>\n",
       "      <td>0.000054</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000069</td>\n",
       "      <td>0.000047</td>\n",
       "      <td>0.000049</td>\n",
       "      <td>0.000126</td>\n",
       "      <td>0.000022</td>\n",
       "      <td>0.000237</td>\n",
       "      <td>0.000043</td>\n",
       "      <td>0.000061</td>\n",
       "      <td>0.000026</td>\n",
       "      <td>0.000058</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>test_2</td>\n",
       "      <td>0.000046</td>\n",
       "      <td>0.000089</td>\n",
       "      <td>0.000035</td>\n",
       "      <td>0.000075</td>\n",
       "      <td>0.000039</td>\n",
       "      <td>0.000036</td>\n",
       "      <td>0.000063</td>\n",
       "      <td>0.000063</td>\n",
       "      <td>0.000055</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000068</td>\n",
       "      <td>0.000048</td>\n",
       "      <td>0.000048</td>\n",
       "      <td>0.000122</td>\n",
       "      <td>0.000021</td>\n",
       "      <td>0.000163</td>\n",
       "      <td>0.000041</td>\n",
       "      <td>0.000065</td>\n",
       "      <td>0.000029</td>\n",
       "      <td>0.000055</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>test_3</td>\n",
       "      <td>0.000042</td>\n",
       "      <td>0.000078</td>\n",
       "      <td>0.000033</td>\n",
       "      <td>0.000078</td>\n",
       "      <td>0.000040</td>\n",
       "      <td>0.000034</td>\n",
       "      <td>0.000057</td>\n",
       "      <td>0.000060</td>\n",
       "      <td>0.000053</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000067</td>\n",
       "      <td>0.000047</td>\n",
       "      <td>0.000046</td>\n",
       "      <td>0.000112</td>\n",
       "      <td>0.000020</td>\n",
       "      <td>0.000154</td>\n",
       "      <td>0.000039</td>\n",
       "      <td>0.000065</td>\n",
       "      <td>0.000027</td>\n",
       "      <td>0.000053</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>test_4</td>\n",
       "      <td>0.000043</td>\n",
       "      <td>0.000083</td>\n",
       "      <td>0.000034</td>\n",
       "      <td>0.000078</td>\n",
       "      <td>0.000041</td>\n",
       "      <td>0.000035</td>\n",
       "      <td>0.000053</td>\n",
       "      <td>0.000061</td>\n",
       "      <td>0.000054</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000067</td>\n",
       "      <td>0.000048</td>\n",
       "      <td>0.000047</td>\n",
       "      <td>0.000122</td>\n",
       "      <td>0.000021</td>\n",
       "      <td>0.000156</td>\n",
       "      <td>0.000039</td>\n",
       "      <td>0.000064</td>\n",
       "      <td>0.000027</td>\n",
       "      <td>0.000054</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 27 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "  utterance   label_0   label_1   label_2   label_3   label_4   label_5  \\\n",
       "0    test_0  0.005376  0.020964  0.008766  0.001087  0.010478  0.004582   \n",
       "1    test_1  0.000051  0.000076  0.000036  0.000105  0.000040  0.000039   \n",
       "2    test_2  0.000046  0.000089  0.000035  0.000075  0.000039  0.000036   \n",
       "3    test_3  0.000042  0.000078  0.000033  0.000078  0.000040  0.000034   \n",
       "4    test_4  0.000043  0.000083  0.000034  0.000078  0.000041  0.000035   \n",
       "\n",
       "    label_6   label_7   label_8  ...  label_16  label_17  label_18  label_19  \\\n",
       "0  0.003384  0.009285  0.770127  ...  0.007280  0.007319  0.010551  0.006741   \n",
       "1  0.000062  0.000067  0.000054  ...  0.000069  0.000047  0.000049  0.000126   \n",
       "2  0.000063  0.000063  0.000055  ...  0.000068  0.000048  0.000048  0.000122   \n",
       "3  0.000057  0.000060  0.000053  ...  0.000067  0.000047  0.000046  0.000112   \n",
       "4  0.000053  0.000061  0.000054  ...  0.000067  0.000048  0.000047  0.000122   \n",
       "\n",
       "   label_20  label_21  label_22  label_23  label_24  label_25  \n",
       "0  0.016093  0.029515  0.012482  0.012054  0.002456  0.008578  \n",
       "1  0.000022  0.000237  0.000043  0.000061  0.000026  0.000058  \n",
       "2  0.000021  0.000163  0.000041  0.000065  0.000029  0.000055  \n",
       "3  0.000020  0.000154  0.000039  0.000065  0.000027  0.000053  \n",
       "4  0.000021  0.000156  0.000039  0.000064  0.000027  0.000054  \n",
       "\n",
       "[5 rows x 27 columns]"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# merge datasets to single frame\n",
    "fullset = pd.concat([df_test, df_calibration], ignore_index=True)\n",
    "assert fullset.utterance.value_counts().max() == 1, \"Hashes/utterances should be unique in full set\"\n",
    "fullset = fullset.drop(['label', 'hash', 'text'], axis=1)\n",
    "fullset.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "5e9b56da-47f0-47a0-813f-2635fa283889",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "# parameters for evaluation\n",
    "X_train = fullset[['utterance']]\n",
    "X_calib = df_calibration[['utterance']]\n",
    "X_test = df_test[['utterance']]\n",
    "\n",
    "a_start = 0.01\n",
    "a_stop = .4\n",
    "a_step = 0.01\n",
    "alpha = np.arange(a_start, a_stop, a_step)\n",
    "\n",
    "max_ps_size = math.ceil(N_INTENTS * .25)\n",
    "max_ps_size = 7\n",
    "\n",
    "target_alpha = .01\n",
    "target_index = index_for_arange(a_start, a_step ,target_alpha)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "7d4c5bc0-fe51-4dfd-aab0-888fd479a40b",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "# instantiate a mock sklearn model which simply looks up the results in a lookup table\n",
    "model = MockModel(fullset, 'utterance', softmax=False, classes=range(N_INTENTS))\n",
    "# fit the test dataset\n",
    "model.fit(X_test, df_test['label'])\n",
    "# extract the test targets from the datasets\n",
    "y_test = df_test['label'].apply(lambda lbl: list(model.classes_).index(lbl))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "2dacf640-6c20-4f5d-82e9-f694a7fc763d",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/floris/miniconda3/envs/cicc/lib/python3.12/site-packages/mapie/classification.py:956: UserWarning: WARNING: your calibration dataset has less labels than your training dataset (training has 26 unique labels while calibration have 17 unique labels\n",
      "  warnings.warn(\n",
      "/Users/floris/Documents/archive/publications/cicc/cicc/src/eval.py:101: UserWarning: Warning: converting a masked element to nan.\n",
      "  return float(agg_width)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(0.985519591141397, 0.9761499148211243, 2.5357142857142856, 0.0)"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# marginal conformal prediction\n",
    "mapie_marginal = MapieClassifier(estimator=model, cv=\"prefit\", method=\"score\", random_state=42)\n",
    "mapie_marginal.fit(X_calib, df_calibration['label'])\n",
    "y_pred_marginal, y_ps_marginal = mapie_marginal.predict(X_test, alpha=alpha)\n",
    "\n",
    "\n",
    "coverages_marginal = [\n",
    "    classification_coverage_score(y_test, y_ps_marginal[:, :, i])\n",
    "    for i, _ in enumerate(alpha)\n",
    "]\n",
    "cqsize_marginal = [\n",
    "    adjusted_classification_x_width_score(y_ps_marginal[:, :, i], np.mean,max_ps_size)\n",
    "    for i, _ in enumerate(alpha)\n",
    "]\n",
    "single_marginal = [\n",
    "    set_size_equals(y_ps_marginal[:,:,i], 1)\n",
    "    for i, _ in enumerate(alpha)\n",
    "]\n",
    "ambiguous_marginal = [\n",
    "    (y_ps_marginal[:, :, i].sum(axis=1) >= max_ps_size).mean()\n",
    "    for i, _ in enumerate(alpha)\n",
    "]\n",
    "ssc_marginal = [\n",
    "    size_stratified_classification_coverage_score(y_test, y_ps_marginal[:, :, i])\n",
    "    for i, _ in enumerate(alpha)\n",
    "]\n",
    "\n",
    "results = cq_stats(y_ps_marginal[:, :, target_index], y_test, max_ps_size)\n",
    "\n",
    "write_results('atis-bert-cicc', dataset, intent_classifier, 'cicc', results)\n",
    "results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "22e7584f-d77c-4fe6-a47d-36d2bad9e77c",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/floris/miniconda3/envs/cicc/lib/python3.12/site-packages/mapie/classification.py:956: UserWarning: WARNING: your calibration dataset has less labels than your training dataset (training has 26 unique labels while calibration have 17 unique labels\n",
      "  warnings.warn(\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(0.996135265700483, 0.7776831345826235, 4.073770491803279, 0.12350936967632027)"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# adaptive conformal prediction\n",
    "mapie_adaptive = MapieClassifier(estimator=model, cv=\"prefit\", method=\"cumulated_score\", random_state=42)\n",
    "mapie_adaptive.fit(X_calib, df_calibration['label'])\n",
    "y_pred_adaptive, y_ps_adaptive = mapie_adaptive.predict(X_test, alpha=alpha)\n",
    "\n",
    "\n",
    "coverages_adaptive = [\n",
    "    classification_coverage_score(y_test, y_ps_adaptive[:, :, i])\n",
    "    for i, _ in enumerate(alpha)\n",
    "]\n",
    "cqsize_adaptive = [\n",
    "    adjusted_classification_x_width_score(y_ps_adaptive[:, :, i], np.mean,max_ps_size)\n",
    "    for i, _ in enumerate(alpha)\n",
    "]\n",
    "single_adaptive = [\n",
    "    set_size_equals(y_ps_adaptive[:,:,i], 1)\n",
    "    for i, _ in enumerate(alpha)\n",
    "]\n",
    "ambiguous_adaptive = [\n",
    "    (y_ps_adaptive[:, :, i].sum(axis=1) >= max_ps_size).mean()\n",
    "    for i, _ in enumerate(alpha)\n",
    "]\n",
    "ssc_adaptive = [\n",
    "    size_stratified_classification_coverage_score(y_test, y_ps_adaptive[:, :, i])\n",
    "    for i, _ in enumerate(alpha)\n",
    "]\n",
    "\n",
    "cq_stats(y_ps_adaptive[:, :, target_index], y_test, max_ps_size)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "b5878c79-c2cb-4d63-a9cd-16961219311c",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/floris/miniconda3/envs/cicc/lib/python3.12/site-packages/mapie/classification.py:956: UserWarning: WARNING: your calibration dataset has less labels than your training dataset (training has 26 unique labels while calibration have 17 unique labels\n",
      "  warnings.warn(\n",
      "/Users/floris/Documents/archive/publications/cicc/cicc/src/eval.py:101: UserWarning: Warning: converting a masked element to nan.\n",
      "  return float(agg_width)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(0.9872231686541738, 0.0, 2.0, 0.0)"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# TOP-K conformal prediction\n",
    "mapie_topk = MapieClassifier(estimator=model, cv=\"prefit\", method=\"top_k\", random_state=42)\n",
    "mapie_topk.fit(X_calib, df_calibration['label'])\n",
    "y_pred_topk, y_ps_topk = mapie_topk.predict(X_test, alpha=alpha)\n",
    "\n",
    "\n",
    "coverages_topk = [\n",
    "    classification_coverage_score(y_test, y_ps_topk[:, :, i])\n",
    "    for i, _ in enumerate(alpha)\n",
    "]\n",
    "cqsize_topk = [\n",
    "    adjusted_classification_x_width_score(y_ps_topk[:, :, i], np.mean,max_ps_size)\n",
    "    for i, _ in enumerate(alpha)\n",
    "]\n",
    "single_topk = [\n",
    "    set_size_equals(y_ps_topk[:,:,i], 1)\n",
    "    for i, _ in enumerate(alpha)\n",
    "]\n",
    "ambiguous_topk = [\n",
    "    (y_ps_topk[:, :, i].sum(axis=1) >= max_ps_size).mean()\n",
    "    for i, _ in enumerate(alpha)\n",
    "]\n",
    "ssc_topk = [\n",
    "    size_stratified_classification_coverage_score(y_test, y_ps_topk[:, :, i])\n",
    "    for i, _ in enumerate(alpha)\n",
    "]\n",
    "\n",
    "cq_stats(y_ps_topk[:, :, target_index], y_test, max_ps_size)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "802ced47-cf35-4648-9dab-49bc0daabe82",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/floris/miniconda3/envs/cicc/lib/python3.12/site-packages/mapie/classification.py:956: UserWarning: WARNING: your calibration dataset has less labels than your training dataset (training has 26 unique labels while calibration have 17 unique labels\n",
      "  warnings.warn(\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(0.9958115183246073, 0.6942078364565588, 3.807142857142857, 0.217206132879046)"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# RAPS\n",
    "mapie_raps = MapieClassifier(estimator=model, cv=\"prefit\", method=\"raps\")\n",
    "mapie_raps.fit(X_calib, df_calibration['label'])\n",
    "y_pred_raps, y_ps_raps = mapie_raps.predict(X_test, alpha=alpha)\n",
    "\n",
    "\n",
    "coverages_raps = [\n",
    "    classification_coverage_score(y_test, y_ps_raps[:, :, i])\n",
    "    for i, _ in enumerate(alpha)\n",
    "]\n",
    "cqsize_raps = [\n",
    "    adjusted_classification_x_width_score(y_ps_raps[:, :, i], np.mean,max_ps_size)\n",
    "    for i, _ in enumerate(alpha)\n",
    "]\n",
    "single_raps = [\n",
    "    set_size_equals(y_ps_raps[:,:,i], 1)\n",
    "    for i, _ in enumerate(alpha)\n",
    "]\n",
    "ambiguous_raps = [\n",
    "    (y_ps_raps[:, :, i].sum(axis=1) >= max_ps_size).mean()\n",
    "    for i, _ in enumerate(alpha)\n",
    "]\n",
    "ssc_raps = [\n",
    "    size_stratified_classification_coverage_score(y_test, y_ps_raps[:, :, i])\n",
    "    for i, _ in enumerate(alpha)\n",
    "]\n",
    "\n",
    "cq_stats(y_ps_raps[:, :, target_index], y_test, max_ps_size)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "bdc5c834",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(0.9948892674616695, 0.7308347529812607, 5.0, 0.0)"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# heuristic 1: select all classes above a certain cutoff or the top-5 if there are none\n",
    "topk_cutoff = TopKCutoffSelector(estimator=model, cutoff=1-target_alpha, k=5)\n",
    "y_ps_h1 = topk_cutoff.predict(X_test)\n",
    "results = cq_stats(y_ps_h1, y_test, max_ps_size)\n",
    "write_results('atis-bert-h1', dataset, intent_classifier, 'h1', results)\n",
    "results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "0cfd2cbd",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/floris/Documents/archive/publications/cicc/cicc/src/eval.py:101: UserWarning: Warning: converting a masked element to nan.\n",
      "  return float(agg_width)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(0.9787052810902896, 1.0, nan, 0.0)"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "top_95 = HeuristicCutoffSelector(estimator=model, cutoff=1-target_alpha)\n",
    "y_ps_h2 = top_95.predict(X_test)\n",
    "results = cq_stats(y_ps_h2, y_test, max_ps_size)\n",
    "write_results('atis-bert-h2', dataset, intent_classifier, 'h2', results)\n",
    "results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "630f3659-2e17-4058-a407-9dba6b9fae52",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(0.9965928449744463, 0.0, 5.0, 0.0)"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "topk_5 = TopKSelector(estimator=model, k=5)\n",
    "y_ps_h3 = topk_5.predict(X_test)\n",
    "\n",
    "results = cq_stats(y_ps_h3, y_test, max_ps_size)\n",
    "write_results('atis-bert-h3', dataset, intent_classifier, 'h3', results)\n",
    "results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "63c93737",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 209.974x449.656 with 3 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot.plot_coverages_widths_multi(\n",
    "        alpha,\n",
    "        [\n",
    "            coverages_marginal,\n",
    "            coverages_adaptive,\n",
    "            coverages_raps,\n",
    "            ],\n",
    "        [\n",
    "            cqsize_marginal,\n",
    "            cqsize_adaptive,\n",
    "            cqsize_raps\n",
    "            ],\n",
    "        ['Marg.', 'Cond.', 'RAPS'],\n",
    "        legend=(2, 'lower left'),\n",
    "        sscs=[\n",
    "            ssc_marginal,\n",
    "            ssc_adaptive,\n",
    "            ssc_raps,\n",
    "        ],\n",
    "        # hlines=(\n",
    "        #     (results_h1[0], ssc_top_5_95, coverages_top_5_95,  'H1', 'dashed'),\n",
    "        #     (widths_mean_top_95, ssc_top_95, coverages_top_95,  'H2', 'dotted'),\n",
    "        #     (widths_mean_topk_5, ssc_topk_5, coverages_topk_5,  'H3', '-.'),\n",
    "        # ),\n",
    "        plot_ssc=True,\n",
    "        markers=('o', 'v', '^',),\n",
    "        dataset=dataset,\n",
    "        model=intent_classifier,\n",
    "        title=\"ATIS\"\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "cf84eaa7-7195-42d2-9803-4fc903b380db",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 303.149x224.828 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "to_plot = [\n",
    "    (y_ps_marginal, 'marginal', '-', None),\n",
    "    (y_ps_adaptive, 'adaptive', '--', None),\n",
    "    (y_ps_raps, 'RAPS', '-.', None),\n",
    "    # (y_ps_topkmapie, 'TOPK', ':', None),\n",
    "    # (y_ps_top_5_95, 'heuristic a', 'dashed', 'black'),\n",
    "    # (y_ps_top_95,  'heuristic b', 'dotted', 'black'),\n",
    "    # (y_ps_topk_5,  'heuristic c', '-.', 'black'),\n",
    "]\n",
    "fig, ax = plt.subplots(1, 1, figsize=plot.fig_dim_1)\n",
    "for (resultset, label, style, color) in to_plot:\n",
    "    if len(label) > len('heuristic') and label[:len('heuristic')] == 'heuristic':\n",
    "        to_plot_result = resultset\n",
    "    else:\n",
    "        to_plot_result = resultset[:, :, target_index]\n",
    "    ax.plot(\n",
    "        np.cumsum([set_size_equals(to_plot_result, i) for i in range(78)]),\n",
    "        label=label,\n",
    "        linestyle=style,\n",
    "        color=color,\n",
    "        linewidth=3,\n",
    "        alpha=.8)\n",
    "    # ax.set_xscale('log')\n",
    "plt.xlabel('Set size <')\n",
    "plt.ylabel('# samples')\n",
    "plt.title(f'{dataset} {intent_classifier}'.format(target_alpha))\n",
    "plt.legend(handlelength=3.5)\n",
    "plt.tight_layout()\n",
    "plt.savefig(f'results/cum_set_size_{dataset}_{intent_classifier}.png', dpi=1200)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "8b7d01bf-b41d-4bbb-b121-44c17cb0a631",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[(19,\n",
       "  'flight_no',\n",
       "  'what is the flight number for the continental flight which leaves denver at 1220 pm and goes to san francisco'),\n",
       " (31, 'city', 'what time zone is denver in'),\n",
       " (39, 'city', 'where is lester pearson airport'),\n",
       " (84, 'flight_no', 'flight number from dallas to houston'),\n",
       " (99,\n",
       "  'ground_fare',\n",
       "  'how much is the ground transportation between atlanta and downtown'),\n",
       " (107,\n",
       "  'flight_time',\n",
       "  'what is the minimum connection time for houston intercontinental'),\n",
       " (276, 'flight', 'us 201'),\n",
       " (279, 'city', 'where is general mitchell international located'),\n",
       " (319, 'airport', 'please list information regarding san francisco airport'),\n",
       " (424,\n",
       "  'aircraft',\n",
       "  \"i need a plane that arrives in san francisco by 9 o'clock pm on friday and leaves from baltimore\"),\n",
       " (501,\n",
       "  'ground_service',\n",
       "  'how long does the ground transportation take from the salt lake city airport'),\n",
       " (549,\n",
       "  'meal',\n",
       "  'what meals are served on american flight 665 673 from milwaukee to seattle'),\n",
       " (553, 'aircraft', 'is there a plane from boston to washington'),\n",
       " (586,\n",
       "  'city',\n",
       "  'show me city served both by nationair and canadian airlines international'),\n",
       " (596,\n",
       "  'day_name',\n",
       "  'what days of the week do flights from san jose to nashville fly on'),\n",
       " (612, 'city', 'what cities does northwest fly to'),\n",
       " (721, 'city', 'where is mco'),\n",
       " (761, 'flight', 'when are the american flights from phoenix to milwaukee'),\n",
       " (874, 'flight_no', 'flight numbers from chicago to seattle on continental'),\n",
       " (882, 'meal', 'what are my meal options from boston to denver'),\n",
       " (890, 'city', 'show me the cities served by canadian airlines international'),\n",
       " (903,\n",
       "  'flight_no',\n",
       "  'which is the flight number for the us air flight from philadelphia to boston is it 279 or is it 137338'),\n",
       " (936,\n",
       "  'flight',\n",
       "  'what are the flights and prices from la to charlotte for monday morning'),\n",
       " (1045,\n",
       "  'flight',\n",
       "  'show me the connecting flights between boston and denver and the types of aircraft used'),\n",
       " (1053, 'city', 'what are the cities that american airlines serves'),\n",
       " (1060, 'city', 'show me all the cities that midwest express serves'),\n",
       " (1114, 'flight', 'i need a reservation from baltimore to san francisco'),\n",
       " (1157,\n",
       "  'flight_no',\n",
       "  'what is the number of first class flights on american airlines')]"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_ps_which = y_ps_marginal\n",
    "ambiguous_indices = np.where((y_ps_which[:, :, target_index].sum(axis=1) >= 2))\n",
    "ambiguous = df_test.loc[ambiguous_indices]\n",
    "list(zip(ambiguous.index,[ labels[l] for l in ambiguous.label.tolist()], ambiguous.text.tolist()))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "fb5e7c35-7990-4491-aa61-016575b8f9fb",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "what is the flight number for the continental flight which leaves denver at 1220 pm and goes to san francisco & flight no & capacity, flight no, flight time, quantity \\\\\n",
      "what time zone is denver in & city & city, distance, flight time \\\\\n",
      "where is lester pearson airport & city & airport, city \\\\\n",
      "flight number from dallas to houston & flight no & capacity, flight+airfare, flight no, flight time, quantity \\\\\n",
      "how much is the ground transportation between atlanta and downtown & ground fare & ground fare, ground service \\\\\n",
      "what is the minimum connection time for houston intercontinental & flight time & distance, flight time, ground fare, ground service \\\\\n",
      "us 201 & flight & abbreviation, distance, ground service \\\\\n",
      "where is general mitchell international located & city & airport, city \\\\\n",
      "please list information regarding san francisco airport & airport & airport, city, ground service \\\\\n",
      "i need a plane that arrives in san francisco by 9 o'clock pm on friday and leaves from baltimore & aircraft & aircraft, flight \\\\\n",
      "how long does the ground transportation take from the salt lake city airport & ground service & distance, ground service \\\\\n",
      "what meals are served on american flight 665 673 from milwaukee to seattle & meal & capacity, meal \\\\\n",
      "is there a plane from boston to washington & aircraft & aircraft, flight \\\\\n",
      "show me city served both by nationair and canadian airlines international & city & airport, city \\\\\n",
      "what days of the week do flights from san jose to nashville fly on & day name & city, flight time \\\\\n",
      "what cities does northwest fly to & city & airport, city \\\\\n",
      "where is mco & city & airport, city \\\\\n",
      "when are the american flights from phoenix to milwaukee & flight & flight, flight time \\\\\n",
      "flight numbers from chicago to seattle on continental & flight no & capacity, flight+airfare, flight no, flight time \\\\\n",
      "what are my meal options from boston to denver & meal & capacity, meal \\\\\n",
      "show me the cities served by canadian airlines international & city & airport, city \\\\\n",
      "which is the flight number for the us air flight from philadelphia to boston is it 279 or is it 137338 & flight no & capacity, flight no, quantity \\\\\n",
      "what are the flights and prices from la to charlotte for monday morning & flight & airfare, flight+airfare \\\\\n",
      "show me the connecting flights between boston and denver and the types of aircraft used & flight & aircraft, flight \\\\\n",
      "what are the cities that american airlines serves & city & airport, city \\\\\n",
      "show me all the cities that midwest express serves & city & airport, city \\\\\n",
      "i need a reservation from baltimore to san francisco & flight & airfare, flight, ground service \\\\\n",
      "what is the number of first class flights on american airlines & flight no & capacity, flight no, quantity \\\\\n"
     ]
    }
   ],
   "source": [
    "for a in ambiguous.index:\n",
    "    alternatives = np.where(y_ps_which[a, :, target_index])[0]\n",
    "    print(\" & \".join([ambiguous.loc[a].text, labels[ambiguous.loc[a].label].replace(\"_\", \" \"), ', '.join([labels[l].replace(\"_\", \" \") for l in alternatives])]) + \" \\\\\\\\\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "b9b26fca-9da4-49f2-b15a-020dfa5623d8",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 606.299x200 with 3 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "label_sorting_map = {i: x for x, i in enumerate(df_train.label.value_counts(sort=True).index)}\n",
    "for i in range(len(labels)):\n",
    "    if i not in label_sorting_map:\n",
    "        label_sorting_map[i] = len(label_sorting_map)\n",
    "fig, axs = plt.subplots(1,3,figsize=(plot.fig_width_in * 2, 2), sharey=True)\n",
    "pd.Series(list(map(lambda x: label_sorting_map[x], df_train.label, ))).hist(bins=N_INTENTS, ax=axs[0])\n",
    "pd.Series(list(map(lambda x: label_sorting_map[x], df_calibration.label, ))).hist(bins=N_INTENTS, ax=axs[1])\n",
    "pd.Series(list(map(lambda x: label_sorting_map[x], df_test.label, ))).hist(bins=N_INTENTS, ax=axs[2])\n",
    "titles = ['Train set', 'Calibration set', 'Test set']\n",
    "[ax.set_xlabel('intent') for ax in axs]\n",
    "[ax.set_title(title) for (ax, title) in zip(axs, titles)]\n",
    "axs[0].set_ylabel('Occurrence')\n",
    "# plt.yticks([]) \n",
    "plt.savefig(f\"results/label_distribution_{dataset}.png\", dpi=1200)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "f1248446",
   "metadata": {},
   "outputs": [],
   "source": [
    "cq_inputs = df_test[cq_filter(y_ps_marginal[:, :, target_index], max_ps_size)].text\n",
    "cq_lbls = cq_labels(y_ps_marginal[:, :, target_index], max_ps_size, labels)\n",
    "\n",
    "cqs = []\n",
    "for i, (text, lbls) in enumerate(zip(cq_inputs, cq_lbls)):\n",
    "    cqs.append({\n",
    "        \"id\": i,\n",
    "        \"text\": text,\n",
    "        \"predictions\": lbls,\n",
    "    })\n",
    "\n",
    "with open(f'results/clarification_questions_{dataset}_{intent_classifier}.json', 'w') as f:\n",
    "    json.dump(cqs, f, indent=4)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "b861e89c",
   "metadata": {},
   "outputs": [],
   "source": [
    "pd.Series(np.cumsum([set_size_equals(y_ps_marginal[:, :, target_index], i) for i in range(N_INTENTS+1)])).to_csv(f'results/cum_{dataset}_{intent_classifier}.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "6f868fec",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.12.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}