{
 "cells": [
  {
   "cell_type": "code",
   "id": "initial_id",
   "metadata": {
    "collapsed": true,
    "ExecuteTime": {
     "end_time": "2025-07-04T17:20:42.362428Z",
     "start_time": "2025-07-04T17:20:42.350473Z"
    }
   },
   "source": [
    "import json\n",
    "\n",
    "import numpy as np\n",
    "from tabulate import tabulate\n",
    "\n",
    "with open(\"./../data/test_performance_output.json\", \"r\") as f:\n",
    "    data = json.load(f)\n",
    "\n",
    "table = []\n",
    "\n",
    "print(f\"Num test results {len(data)}\")\n",
    "\n",
    "for test in data:\n",
    "    # print(f\"Test: {test['input']}\")\n",
    "    result = test['result']\n",
    "    num_expected = len(test['expected'])\n",
    "    num_terms = len(result[\"terms\"])\n",
    "    num_definitions = len([definition for term in result[\"terms\"] for definition in term[\"definitions\"]])\n",
    "\n",
    "    true_expected = test['trueExpected']\n",
    "    allowed_expected = test['allowedExpected']\n",
    "    allowed_unexpected = test['allowedUnexpected']\n",
    "    false_observed = test['falseObserved']\n",
    "    false_definitions = test['falseDefinitions']\n",
    "\n",
    "    table.append([\n",
    "        1 if num_expected == 0 else true_expected / num_expected,\n",
    "        1 if num_expected == 0 else allowed_expected / num_expected,\n",
    "        1 if num_terms == 0 else (num_terms - false_observed) / num_terms,\n",
    "        1 if num_definitions == 0 else (num_definitions - false_definitions) / num_definitions\n",
    "    ])\n",
    "\n",
    "print(tabulate(table, headers=[\n",
    "    \"Recall (Strict)\",\n",
    "    \"Recall\",\n",
    "    \"Precision\",\n",
    "    \"Gen. Def. Validity\"\n",
    "]))\n",
    "\n",
    "print(\"\\n#### Averages ####\\n\")\n",
    "\n",
    "print(tabulate([np.average(np.array(table), axis=0)], headers=[\n",
    "    \"Recall (Strict)\",\n",
    "    \"Recall\",\n",
    "    \"Precision\",\n",
    "    \"Gen. Def. Validity\"\n",
    "]))\n",
    "\n",
    "# data"
   ],
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Num test results 21\n",
      "  Recall (Strict)    Recall    Precision    Gen. Def. Validity\n",
      "-----------------  --------  -----------  --------------------\n",
      "         0.666667  1            1                     1\n",
      "         1         1            0.8                   0.25\n",
      "         0.666667  1            0.75                  1\n",
      "         1         1            0.6                   1\n",
      "         1         1            0.8                   0.333333\n",
      "         0.714286  0.857143     1                     0.5\n",
      "         0.8       1            1                     0.5\n",
      "         1         1            0                     1\n",
      "         1         1            1                     1\n",
      "         1         1            0.6                   1\n",
      "         1         1            0.428571              1\n",
      "         1         1            1                     1\n",
      "         1         1            1                     1\n",
      "         1         1            1                     0\n",
      "         0.8       1            1                     1\n",
      "         1         1            0.666667              1\n",
      "         0.75      0.75         0.666667              1\n",
      "         1         1            0.333333              1\n",
      "         1         1            1                     1\n",
      "         1         1            1                     1\n",
      "         0.8       1            0.833333              1\n",
      "\n",
      "#### Averages ####\n",
      "\n",
      "  Recall (Strict)    Recall    Precision    Gen. Def. Validity\n",
      "-----------------  --------  -----------  --------------------\n",
      "         0.914172  0.981293     0.784694              0.837302\n"
     ]
    }
   ],
   "execution_count": 63
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}