{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "2e1f00d3-5184-41ad-b263-2731bd6907f0",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "from Utils import calcular_resultados"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "d681e1a6-c967-4249-a70d-c6e412175824",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "dataset = pd.read_csv('dataset//Teste_JBCS25.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "d7b27a05-0583-4911-9174-ddfe0583446d",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "def selecionar_index(dataset):\n",
    "    index_a = []\n",
    "    index_b = []\n",
    "    for index, row in dataset.iterrows():\n",
    "        if row['reference'] == 'grader_a':\n",
    "            index_a.append(index)\n",
    "        elif row['reference'] == 'grader_b':\n",
    "            index_b.append(index)\n",
    "        else:\n",
    "            print(\"Erro\")\n",
    "    assert len(index_a) == len(index_b)\n",
    "    return index_a, index_b\n",
    "index_a, index_b = selecionar_index(dataset)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "5729f104-bce9-454e-b348-d6590c52fc44",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0, 1, 5, 7, 8, 11, 12, 14, 15, 17, 19, 21, 23, 25, 26, 27, 29, 33, 38, 39, 43, 44, 45, 46, 48, 49, 52, 53, 56, 57, 58, 59, 61, 62, 65, 67, 69, 70, 73, 75, 79, 82, 86, 91, 92, 94, 96, 98, 99, 100, 101, 102, 103, 105, 108, 113, 114, 117, 118, 119, 122, 123, 125, 126, 128, 133, 135, 136, 137]\n",
      "[2, 3, 4, 6, 9, 10, 13, 16, 18, 20, 22, 24, 28, 30, 31, 32, 34, 35, 36, 37, 40, 41, 42, 47, 50, 51, 54, 55, 60, 63, 64, 66, 68, 71, 72, 74, 76, 77, 78, 80, 81, 83, 84, 85, 87, 88, 89, 90, 93, 95, 97, 104, 106, 107, 109, 110, 111, 112, 115, 116, 120, 121, 124, 127, 129, 130, 131, 132, 134]\n"
     ]
    }
   ],
   "source": [
    "print(index_a)\n",
    "print(index_b)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "4df02146-061a-4594-8470-b582385aedbf",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "a3fcc7d5-29ff-4c8f-8542-c58c114a233e",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "e100fb4a-2b3e-47a1-a01c-14e05a8e8ef8",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "def quebrar_respostas(y, y_hat, index_a, index_b):\n",
    "    y_hat_a = []\n",
    "    y_hat_b = []\n",
    "    y_a, y_b = [], []\n",
    "    #print(y)\n",
    "    for idx in range(len(y)):\n",
    "        if idx in index_a:\n",
    "            y_a.append( y[idx] )\n",
    "            y_hat_a.append( y_hat[idx] )\n",
    "        elif idx in index_b:\n",
    "            y_b.append( y[idx] )\n",
    "            y_hat_b.append( y_hat[idx] )\n",
    "        elif (idx in index_a) and (idx in index_b):\n",
    "            print(\"Erro\")\n",
    "    return y_a, y_b, y_hat_a, y_hat_b\n",
    "\n",
    "def formatar_respostas(dic_a, dic_b, dic_geral):\n",
    "    F1_m_a = sum(dic_a['F1-Macro'])/len(dic_a['F1-Macro'])\n",
    "    F1_w_a = sum(dic_a['F1-Weighted'])/len(dic_a['F1-Weighted'])\n",
    "    QWK_a = sum(dic_a['QWK'])/len(dic_a['QWK'])\n",
    "    print(f\"A: {F1_m_a:.2f} & {F1_w_a:.2f} & {QWK_a:.2f} &\")\n",
    "    F1_m_b = sum(dic_b['F1-Macro'])/len(dic_b['F1-Macro'])\n",
    "    F1_w_b = sum(dic_b['F1-Weighted'])/len(dic_b['F1-Weighted'])\n",
    "    QWK_b = sum(dic_b['QWK'])/len(dic_b['QWK'])\n",
    "    print(f\"B: {F1_m_b:.2f} & {F1_w_b:.2f} & {QWK_b:.2f} &\")\n",
    "    F1_m_geral = sum(dic_geral['F1-Macro'])/len(dic_geral['F1-Macro'])\n",
    "    F1_w_geral = sum(dic_geral['F1-Weighted'])/len(dic_geral['F1-Weighted'])\n",
    "    QWK_geral = sum(dic_geral['QWK'])/len(dic_geral['QWK'])\n",
    "    print(f\"geral: {F1_m_geral:.2f} & {F1_w_geral:.2f} & {QWK_geral:.2f} &\")\n",
    "    F1_m_media = (F1_m_a + F1_m_b) / 2\n",
    "    F1_w_media = (F1_w_a + F1_w_b) / 2 \n",
    "    QWK_media =  (QWK_a + QWK_b) / 2\n",
    "    print(f\"media: {F1_m_media:.2f} & {F1_w_media:.2f} & {QWK_media:.2f} &\")\n",
    "\n",
    "def adicionar_performances(dic_a, dic_b, dic_geral, perf_a, perf_b, perf_geral):\n",
    "    for key in perf_a:\n",
    "        perf_a[key].append(dic_a[key])\n",
    "        perf_b[key].append(dic_b[key])\n",
    "        perf_geral[key].append(dic_geral[key])\n",
    "    \n",
    "def calcular_media(dataset, index_a, index_b):\n",
    "    perf_a = {'QWK': [], 'F1-Macro': [], 'F1-Weighted': []}\n",
    "    perf_b = {'QWK': [], 'F1-Macro': [], 'F1-Weighted': []}\n",
    "    perf_geral = {'QWK': [], 'F1-Macro': [], 'F1-Weighted': []}\n",
    "    for idx, row in dataset.iterrows():\n",
    "        y = eval(row['y'])\n",
    "        y_hat = eval(row['y_hat'])\n",
    "        y_a, y_b, y_hat_a, y_hat_b = quebrar_respostas(y, y_hat, index_a, index_b)\n",
    "        dic_a = calcular_resultados(y_a, y_hat_a)\n",
    "        dic_b = calcular_resultados(y_b, y_hat_b)\n",
    "        dic_geral = calcular_resultados(y_a + y_b, y_hat_a + y_hat_b)\n",
    "        adicionar_performances(dic_a, dic_b, dic_geral, perf_a, perf_b, perf_geral)\n",
    "    formatar_respostas(perf_a, perf_b, perf_geral)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "712a4429-a014-4991-88a1-e69d4cc0441c",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'\\nfor comp in range(1, 6):\\n    print(f\"Comp {comp}\")\\n    respostas = pd.read_csv(f\\'RF-{comp}.csv\\')\\n    calcular_media(respostas, index_a, index_b)\\n'"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "\"\"\"\n",
    "for comp in range(1, 6):\n",
    "    print(f\"Comp {comp}\")\n",
    "    respostas = pd.read_csv(f'RF-{comp}.csv')\n",
    "    calcular_media(respostas, index_a, index_b)\n",
    "\"\"\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "249c4d4b-8e29-43f6-a700-ba20c2b79c07",
   "metadata": {},
   "outputs": [],
   "source": [
    "def calcular_media(dataset, index_a, index_b, nome):\n",
    "    global Y_HAT, NORMALIZACAO\n",
    "    perf_a = {'QWK': [], 'F1-Macro': [], 'F1-Weighted': []}\n",
    "    perf_b = {'QWK': [], 'F1-Macro': [], 'F1-Weighted': []}\n",
    "    perf_geral = {'QWK': [], 'F1-Macro': [], 'F1-Weighted': []}\n",
    "    y = (dataset['label']*NORMALIZACAO).tolist()\n",
    "    y_hat = (dataset[Y_HAT]*NORMALIZACAO).tolist()\n",
    "    y_a, y_b, y_hat_a, y_hat_b = quebrar_respostas(y, y_hat, index_a, index_b)\n",
    "    dic_a = calcular_resultados(y_a, y_hat_a)\n",
    "    dic_b = calcular_resultados(y_b, y_hat_b)\n",
    "    dic_geral = calcular_resultados(y_a + y_b, y_hat_a + y_hat_b)\n",
    "    adicionar_performances(dic_a, dic_b, dic_geral, perf_a, perf_b, perf_geral)\n",
    "    formatar_respostas(perf_a, perf_b, perf_geral, nome)\n",
    "\n",
    "\n",
    "\n",
    "def printar_dic(dic, nome):\n",
    "    global dic_print_global\n",
    "    F1_m = sum(dic['F1-Macro'])/len(dic['F1-Macro'])\n",
    "    F1_w = sum(dic['F1-Weighted'])/len(dic['F1-Weighted'])\n",
    "    QWK = sum(dic['QWK'])/len(dic['QWK'])\n",
    "    print_A = f\"{F1_m:.2f} & {F1_w:.2f} & {QWK:.2f} &\"\n",
    "    print(f\"{nome}: {print_A}\")\n",
    "    dic_print_global[nome] += print_A\n",
    "\n",
    "def fazer_media(dic1, dic2):\n",
    "    F1_macro = (dic1['F1-Macro'][0]+dic2['F1-Macro'][0])/2\n",
    "    F1_weighted = (dic1['F1-Weighted'][0]+dic2['F1-Weighted'][0])/2\n",
    "    #print(dic1['QWK'], dic2['QWK'])\n",
    "    QWK = (dic1['QWK'][0]+dic2['QWK'][0])/2\n",
    "    return {'F1-Macro': [F1_macro], 'F1-Weighted': [F1_weighted], 'QWK': [QWK]}\n",
    "\n",
    "def formatar_respostas(dic_a, dic_b, dic_geral, nome):\n",
    "    global dic_print_global\n",
    "    printar_dic(dic_a, f\"{nome}-A\")\n",
    "    printar_dic(dic_b, f\"{nome}-B\")\n",
    "    printar_dic(dic_geral, f\"{nome}-geral\")\n",
    "    printar_dic(fazer_media(dic_a, dic_b), f\"{nome}-media\")\n",
    "    \"\"\"\n",
    "    print(f\"A: {print_A})\n",
    "    dic_print_global[f\"{nome}-A\"\n",
    "    F1_m_b = sum(dic_b['F1-Macro'])/len(dic_b['F1-Macro'])\n",
    "    F1_w_b = sum(dic_b['F1-Weighted'])/len(dic_b['F1-Weighted'])\n",
    "    QWK_b = sum(dic_b['QWK'])/len(dic_b['QWK'])\n",
    "    print(f\"B: {F1_m_b:.2f} & {F1_w_b:.2f} & {QWK_b:.2f} &\")\n",
    "    F1_m_geral = sum(dic_geral['F1-Macro'])/len(dic_geral['F1-Macro'])\n",
    "    F1_w_geral = sum(dic_geral['F1-Weighted'])/len(dic_geral['F1-Weighted'])\n",
    "    QWK_geral = sum(dic_geral['QWK'])/len(dic_geral['QWK'])\n",
    "    print(f\"geral: {F1_m_geral:.2f} & {F1_w_geral:.2f} & {QWK_geral:.2f} &\")\n",
    "    F1_m_media = (F1_m_a + F1_m_b) / 2\n",
    "    F1_w_media = (F1_w_a + F1_w_b) / 2 \n",
    "    QWK_media =  (QWK_a + QWK_b) / 2\n",
    "    print(f\"media: {F1_m_media:.2f} & {F1_w_media:.2f} & {QWK_media:.2f} &\")\n",
    "    \"\"\"\n",
    "\n",
    "dic_print_global = {}\n",
    "def gerar_nome(modelo, comp):\n",
    "    global dic_print_global\n",
    "    string = \"\"\n",
    "    if modelo == 'bertimbau-base':\n",
    "        string = f\"jbcs2025_bertimbau_base-C{comp}-encoder_classification-C{comp}_inference_results.jsonl\"\n",
    "    elif modelo == 'mbert':\n",
    "        string = f'jbcs2025_mbert_base-C{comp}-encoder_classification-C{comp}_inference_results.jsonl'\n",
    "    elif modelo == 'bertimbau-large':\n",
    "        string = f\"jbcs2025_bertimbau-large-C{comp}-encoder_classification-C{comp}_inference_results.jsonl\"\n",
    "    elif modelo == \"Llama-essay_only\":\n",
    "        string = f\"jbcs2025_llama31_8b-balanced-C{comp}-llama31_classification_lora-C{comp}-essay_only_inference_results.jsonl\"\n",
    "    elif modelo == \"Llama-full_context\":\n",
    "        string = f\"jbcs2025_Llama-3.1-8B-llama31_classification_lora-C{comp}-full_context-llama31_classification_lora-C{comp}-full_context_inference_results.jsonl\"\n",
    "    elif modelo == \"Phi4-essay_only\":\n",
    "        string = f\"jbcs2025_phi4-balanced-C{comp}-phi4_classification_lora-C{comp}-essay_only_inference_results.jsonl\"\n",
    "    elif modelo == \"Phi4-full_context\":\n",
    "        string = f\"jbcs2025_phi-4-phi4_classification_lora-C{comp}-full_context-phi4_classification_lora-C{comp}-full_context_inference_results.jsonl\"\n",
    "    elif modelo == \"Phi3.5-essay-only\":\n",
    "        string = f\"jbcs2025_phi35-balanced-C{comp}-phi35_classification_lora-C{comp}-essay_only_inference_results.jsonl\"\n",
    "    elif modelo == \"Phi3.5-full_context\":\n",
    "        string = f\"jbcs2025_Phi-3.5-mini-instruct-phi35_classification_lora-C{comp}-full_context-phi35_classification_lora-C{comp}-full_context_inference_results.jsonl\"\n",
    "    #sabias\n",
    "    elif modelo == \"sabia-essay-only\":\n",
    "        string = f\"sabia-3-zero-shot-C{comp}-essay_only_inference_results.jsonl\"\n",
    "    elif modelo == \"sabia-full\":\n",
    "        string = f\"sabia-3-zero-shot-C{comp}-full_context_inference_results.jsonl\"\n",
    "    elif modelo == \"sabia-grader-full\":\n",
    "        string = f\"sabia-3-grader-zero-shot-C{comp}-full_context_inference_results.jsonl\"\n",
    "    elif modelo == \"sabia-grader-essay-only\":\n",
    "        string = f\"sabia-3-grader-zero-shot-C{comp}-essay_only_inference_results.jsonl\"\n",
    "    elif modelo == \"sabia-extractor-full\":\n",
    "        string = f\"sabia-3-extractor-zero-shot-C{comp}-full_context_inference_results.jsonl\"\n",
    "    #gpts\n",
    "    elif modelo == 'gpt4-essay-only':\n",
    "        string = f\"gpt-4o-2024-11-20-zero-shot-C{comp}-essay_only_inference_results.jsonl\"\n",
    "    elif modelo == 'gpt4-full':\n",
    "        string = f\"gpt-4o-2024-11-20-zero-shot-C{comp}-full_context_inference_results.jsonl\"\n",
    "    elif modelo == 'gpt4-grader-full':\n",
    "        string = f\"gpt-4o-2024-11-20-zero-shot-grader-C{comp}-full_context_inference_results.jsonl\"\n",
    "    elif modelo == 'gpt4-grader-essay-only':\n",
    "        string = f\"gpt-4o-2024-11-20-grader-zero-shot-C{comp}-essay_only_inference_results.jsonl\"\n",
    "    #deepseeks\n",
    "    elif modelo == 'deepseek-essay-only':\n",
    "        string = f\"deepseek-reasoner-zero-shot-C{comp}-essay_only_inference_results.jsonl\"\n",
    "    elif modelo == 'deepseek-full':\n",
    "        string = f\"deepseek-reasoner-zero-shot-C{comp}-full_context_inference_results.jsonl\"\n",
    "    elif modelo == 'deepseek-grader-essay-only':\n",
    "        string = f\"deepseek-reasoner-grader-zero-shot-C{comp}-essay_only_inference_results.jsonl\"\n",
    "    elif modelo == \"deepseek-grader-full-context\":\n",
    "        string = f\"deepseek-reasoner-grader-zero-shot-C{comp}-full_context_inference_results.jsonl\"\n",
    "    elif modelo == \"deepseek-extractor-full-context\":\n",
    "        string = f\"deepseek-reasoner-extractor-zero-shot-C{comp}-full_context_inference_results.jsonl\"\n",
    "    if comp == 1:\n",
    "        dic_print_global[f\"{modelo}-A\"] = \"\"\n",
    "        dic_print_global[f\"{modelo}-B\"] = \"\"\n",
    "        dic_print_global[f\"{modelo}-geral\"] = \"\"\n",
    "        dic_print_global[f\"{modelo}-media\"] = \"\"\n",
    "    return string"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "cebff6ca-2b47-4960-a155-11da9342d707",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sabia-extractor-full-A: 0.09 & 0.24 & 0.29 &\n",
      "sabia-extractor-full-B: 0.11 & 0.28 & 0.33 &\n",
      "sabia-extractor-full-geral: 0.09 & 0.26 & 0.31 &\n",
      "sabia-extractor-full-media: 0.10 & 0.26 & 0.31 &\n",
      "sabia-extractor-full-A: 0.21 & 0.34 & 0.29 &\n",
      "sabia-extractor-full-B: 0.12 & 0.14 & 0.27 &\n",
      "sabia-extractor-full-geral: 0.15 & 0.21 & 0.28 &\n",
      "sabia-extractor-full-media: 0.16 & 0.24 & 0.28 &\n",
      "sabia-extractor-full-A: 0.20 & 0.37 & 0.24 &\n",
      "sabia-extractor-full-B: 0.20 & 0.21 & 0.19 &\n",
      "sabia-extractor-full-geral: 0.19 & 0.27 & 0.21 &\n",
      "sabia-extractor-full-media: 0.20 & 0.29 & 0.21 &\n",
      "sabia-extractor-full-A: 0.19 & 0.42 & 0.35 &\n",
      "sabia-extractor-full-B: 0.18 & 0.35 & 0.40 &\n",
      "sabia-extractor-full-geral: 0.17 & 0.38 & 0.38 &\n",
      "sabia-extractor-full-media: 0.18 & 0.38 & 0.38 &\n",
      "sabia-extractor-full-A: 0.22 & 0.25 & 0.46 &\n",
      "sabia-extractor-full-B: 0.21 & 0.23 & 0.56 &\n",
      "sabia-extractor-full-geral: 0.23 & 0.23 & 0.52 &\n",
      "sabia-extractor-full-media: 0.22 & 0.24 & 0.51 &\n"
     ]
    }
   ],
   "source": [
    "#print(f\"Comp C1\")\n",
    "modelo = 'sabia-extractor-full'\n",
    "Y_HAT = \"pontuacao\"\n",
    "NORMALIZACAO = 1\n",
    "for comp in range(1, 6):\n",
    "    #print(comp)\n",
    "    nome = gerar_nome(modelo, comp)\n",
    "    respostas = pd.read_json(nome, lines=True)\n",
    "    #print(respostas.head())\n",
    "    calcular_media(respostas, index_a, index_b, modelo)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "3fa030a0-ea8f-4315-a153-6b4f5dfff3b8",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sabia-extractor-full-A & 0.09 & 0.24 & 0.29 &0.21 & 0.34 & 0.29 &0.20 & 0.37 & 0.24 &0.19 & 0.42 & 0.35 &0.22 & 0.25 & 0.46 \\\\\n",
      "sabia-extractor-full-B & 0.11 & 0.28 & 0.33 &0.12 & 0.14 & 0.27 &0.20 & 0.21 & 0.19 &0.18 & 0.35 & 0.40 &0.21 & 0.23 & 0.56 \\\\\n",
      "sabia-extractor-full-geral & 0.09 & 0.26 & 0.31 &0.15 & 0.21 & 0.28 &0.19 & 0.27 & 0.21 &0.17 & 0.38 & 0.38 &0.23 & 0.23 & 0.52 \\\\\n",
      "sabia-extractor-full-media & 0.10 & 0.26 & 0.31 &0.16 & 0.24 & 0.28 &0.20 & 0.29 & 0.21 &0.18 & 0.38 & 0.38 &0.22 & 0.24 & 0.51 \\\\\n"
     ]
    }
   ],
   "source": [
    "for key in dic_print_global:\n",
    "    print(f\"{key} & {dic_print_global[key][:-2] } \\\\\\\\\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "4a60f717-93de-4b80-907a-d1ae06f7db63",
   "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": "3.9.19"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}