demo/BERT/BERT_TUTORIAL.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "BERT_TUTORIAL.ipynb",
      "provenance": [],
      "collapsed_sections": [
        "rJyT8Cp3hb99",
        "qJ-e2fKusHpj",
        "i6hRrVVjAiEB",
        "2vDvFQfyRO1U",
        "YKswmOITHI9L",
        "QdpuvKK_QLtq",
        "2TbkubgI9Br6"
      ],
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/felipemaiapolo/legalnlp/blob/main/demo/BERT/BERT_TUTORIAL.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "dh1lExvDZ_S7"
      },
      "source": [
        "# BERT\n",
        "\n",
        "**Autores**: \n",
        "- Gabriel Caiaffa (gabrielcaiaffa08@gmail.com)\n",
        "- Kauê Capellato J. Parreira (kauecapellatojp@gmail.com)\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "yR7XFCj423Tj"
      },
      "source": [
        "# Inicialização"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "979zktVh2gjb"
      },
      "source": [
        "# inicializando a seed e importando algumas bibliotecas\n",
        "random_seed=42\n",
        "\n",
        "import numpy as np\n",
        "import random\n",
        "\n",
        "np.random.seed(random_seed)\n",
        "random.seed(random_seed)"
      ],
      "execution_count": 1,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "86buPegRZ5z_"
      },
      "source": [
        "import IPython\n",
        "from IPython.display import Image\n",
        "from IPython.display import clear_output"
      ],
      "execution_count": 2,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "M-owYaP70mSQ"
      },
      "source": [
        "# função para exibir as imagens\n",
        "def show(link,width):\n",
        "   return IPython.display.Image(data = link, width = width)"
      ],
      "execution_count": 3,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FcWsY0-BaAVH"
      },
      "source": [
        "# Introdução\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "XcuhPn6AaEuv"
      },
      "source": [
        "Com o amadurecimento das Redes Neurais Recorrentes (RNNs, em inglês), veio o surgimento e desenvolvimento de modelos de processamento de linguagem natural (PLN) mais complexos, capazes de compreender cada vez mais sentenças e seus contextos em um conjunto de textos. Como exemplo de uma aplicação dessas redes, temos o **BERT**, que é um modelo de aprendizado profundo bastante utilizado em diversas tarefas de processamento de linguagem natural.\n",
        "\n",
        "BERT significa **B**idirectional **E**ncoder **R**epresentations for **T**ransformers e foi desenvolvido por pesquisadores do $\\textit{Google AI Language}$ em 2018 e apresentado ao público em 2019, obtendo resultados espetaculares, conforme seu artigo original [[1]](https://arxiv.org/pdf/1810.04805.pdf). Além disso, os códigos do modelo também foram liberados pela equipe e podem ser encontrados neste [Github](https://github.com/google-research/bert).\n",
        "\n",
        "Sendo utilizado ainda como inspiração em diversas arquiteturas de PLN, formas de treinamento e modelos de linguagem natural, como XLNet, ERNIE2.0, RoBERTa, entre outros."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qsN42LAwbnW3"
      },
      "source": [
        "A parte teórica foi inspirada conforme apresentado no artigo original [[1]](https://arxiv.org/pdf/1810.04805.pdf), assim como em [[2]](https://arxiv.org/pdf/1706.03762.pdf), [[3]]( https://nlp.stanford.edu/seminar/details/jdevlin.pdf), [[4]](https://jalammar.github.io/illustrated-bert/), [[5]](https://towardsdatascience.com/bert-explained-state-of-the-art-language-model-for-nlp-f8b21a9b6270) e demonstrações dadas em [[6]](https://colab.research.google.com/drive/1pTuQhug6Dhl9XalKB0zUGf4FIdYFlpcX#scrollTo=EKOTlwcmxmej) e [[7]](https://mccormickml.com/2020/03/10/question-answering-with-a-fine-tuned-BERT/)."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Am1f10iCaVoT"
      },
      "source": [
        "##O que é o BERT e como funciona"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "0vFz-PqOSWq7"
      },
      "source": [
        "O que faz o BERT ser um modelo que se destaca é o fato dele utilizar um treinamento bidirecional do Transformer, que é um tipo de mecanismo de atenção muito efieciente que aprende as relações entre as componentes de um determinado texto.\n",
        "\n",
        "Os Transformers funcionam com dois mecanismos: o $\\textit{encoder}$ e o $\\textit{decoder}$. Basicamente, o encoder funciona transformando o input recebido em contexto e o decoder funciona transformando esse contexto em algum outro objetivo, como por exemplo a tradução para uma outra língua.\n",
        "\n",
        "Para mais informações sobre os Tranformers, leia o artigo original em que foi apresentado [[2]](https://arxiv.org/pdf/1706.03762.pdf).\n",
        "\n",
        "\n",
        "Diferentemente dos modelos unidirecionais baseados em \"contexto livre\", como é o caso do $\\textit{word2vec}$, gerando um $\\textit{embedding}$ (representação do espaço das palavras no espaço real) igual para uma determinada palavra independentemente do seu contexto, o BERT, por ser um modelo bidirecional, consegue extrair o significado da palavra em cada contexto, por isso é chamado também de um \"modelo contextual\".\n",
        "\n",
        "Para exemplificar o parágrafo acima, pense na palavra \"banco\". Nos modelos livres de contexto, essa palavra teria a mesma representação independentemente do seu contexto, embora banco representando uma agência bancária é diferente do banco de uma praça, por exemplo. Por sua vez, os modelos baseados em contexto, identificam cada palavra em seu contexto.\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 270
        },
        "id": "bJKjE3Qyz5Os",
        "outputId": "f55db095-7353-45a4-a789-f300f693cc75"
      },
      "source": [
        "show('https://raw.githubusercontent.com/felipemaiapolo/legalnlp/main/demo/BERT/notebook_images/image1.jpg',600)"
      ],
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "image/jpeg": "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\n",
            "text/plain": [
              "<IPython.core.display.Image object>"
            ]
          },
          "metadata": {
            "image/jpeg": {
              "width": 600
            }
          },
          "execution_count": 4
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "TPKlod_VNlqb"
      },
      "source": [
        "Fonte: Imagem adaptada de [[3]](https://nlp.stanford.edu/seminar/details/jdevlin.pdf)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 262
        },
        "id": "B_cxmYXw0VBe",
        "outputId": "84d903e7-2981-40c7-d45d-5ab4e7c7a7ad"
      },
      "source": [
        "show('https://raw.githubusercontent.com/felipemaiapolo/legalnlp/main/demo/BERT/notebook_images/image2.jpg',600)"
      ],
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "image/jpeg": "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\n",
            "text/plain": [
              "<IPython.core.display.Image object>"
            ]
          },
          "metadata": {
            "image/jpeg": {
              "width": 600
            }
          },
          "execution_count": 5
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Ia2EkCw1A-ex"
      },
      "source": [
        "Fonte: Imagem adaptada de [[3]](https://nlp.stanford.edu/seminar/details/jdevlin.pdf)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_MhPO3jM0M4K"
      },
      "source": [
        "Além disso, o BERT foi treinado utilizando uma enorme quantidade de dados não rotulados da Wikipedia (cerca de 2,5 bilhões de palavras) e em corpus de livros (cerca de 800 milhões de palavras).\n",
        "\n",
        "Agora veremos brevemente a arquitetura desse modelo."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bDJG3cafQ51J"
      },
      "source": [
        "##Arquitetura"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "BoRcRARcQ8EQ"
      },
      "source": [
        "Quanto à arquitetura, o BERT possui dois tipos, de acordo com o artigo citado acima: o Base e o Large. As diferenças estão listadas abaixo:\n",
        "\n",
        "**BERTBASE:** \n",
        "\n",
        "  * L = 12 \\\n",
        "  * H = 768 \\\n",
        "  * A = 12 \\\n",
        "  * Total de Parâmetros = 110 milhões\n",
        "\n",
        "**BERTLARGE:**\n",
        "\n",
        "  * L = 24 \\\n",
        "  * H = 1024 \\\n",
        "  * A = 16 \\\n",
        "  * Total de Parâmetros = 340 milhões\n",
        "\n",
        "\\\n",
        "**Onde**:\\\n",
        "    L = Número de camadas (Blocos Transformer)\\\n",
        "    H = Quantidade de unidades na rede neural\\\n",
        "    A = Cabeças de auto-atenção"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 486
        },
        "id": "o-QnGtbW0gxB",
        "outputId": "95df5154-181f-47d2-acd6-c5d90d322340"
      },
      "source": [
        "show('https://raw.githubusercontent.com/felipemaiapolo/legalnlp/main/demo/BERT/notebook_images/bert1.jpg',600)"
      ],
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "image/jpeg": "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\n",
            "text/plain": [
              "<IPython.core.display.Image object>"
            ]
          },
          "metadata": {
            "image/jpeg": {
              "width": 600
            }
          },
          "execution_count": 6
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "RjkhjcW6PBED"
      },
      "source": [
        "Fonte: Imagem adaptada de [[4]](https://jalammar.github.io/illustrated-bert/)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "rJyT8Cp3hb99"
      },
      "source": [
        "### Entrada do Modelo"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "T-DHUwyxhePA"
      },
      "source": [
        "A entrada do modelo pode ser composta por uma sentença ou por um par de sentenças, com tamanho máximo suportado pelo modelo de 512 tokens. O BERT utiliza a tokenização WordPiece.\n",
        "\n",
        "O primeiro token de entrada é sempre marcado como [CLS], que é um token especial de classificação. Quando duas frases entram no modelo, elas podem ser diferenciadas por meio do token [SEP]."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 426
        },
        "id": "18U5ahxS2QUN",
        "outputId": "a30379ac-1fd3-486f-86bf-c11a292de395"
      },
      "source": [
        "show('https://raw.githubusercontent.com/felipemaiapolo/legalnlp/main/demo/BERT/notebook_images/enconder1.jpg',600)"
      ],
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "image/jpeg": "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\n",
            "text/plain": [
              "<IPython.core.display.Image object>"
            ]
          },
          "metadata": {
            "image/jpeg": {
              "width": 600
            }
          },
          "execution_count": 7
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "1AuU4m4-PDOd"
      },
      "source": [
        "Fonte: Imagem adaptada de [[4]](https://jalammar.github.io/illustrated-bert/)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 335
        },
        "id": "QhlgJxkx2mWv",
        "outputId": "7e55b9b8-9d84-41cd-c2c2-761f8ba4b739"
      },
      "source": [
        "show('https://raw.githubusercontent.com/felipemaiapolo/legalnlp/main/demo/BERT/notebook_images/enconder2.jpg',600)"
      ],
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "image/jpeg": "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\n",
            "text/plain": [
              "<IPython.core.display.Image object>"
            ]
          },
          "metadata": {
            "image/jpeg": {
              "width": 600
            }
          },
          "execution_count": 8
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6JyP8sX-PFlD"
      },
      "source": [
        "Fonte: Imagem adaptada de [[5]](https://towardsdatascience.com/bert-explained-state-of-the-art-language-model-for-nlp-f8b21a9b6270)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "fy-GA4lu14Zu"
      },
      "source": [
        "Para um dado token de entrada, sua forma é dada por um embedding da soma de 3 outros tokens, como visto na figura acima. São eles:\n",
        "\n",
        "* **Position Embeddings**: O modelo aprende e utiliza embeddings de posição para expressar a posição/ordem das palavras na sentença. \n",
        "\n",
        "* **Segment Embeddings**: O BERT recebe pares de sentenças e utiliza embeddings para o modelo aprender e conseguir distingui-las.  \n",
        "\n",
        "* **Token Embeddings**: São embeddings específicos, aprendidos através do vocabulário do WordPiece."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qJ-e2fKusHpj"
      },
      "source": [
        "### Saída do Modelo"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "1aHs8X5ZsLlC"
      },
      "source": [
        "Já a saída do modelo consiste em um vetor de tamanho $\\textit{hidden_size}$, que no caso do BERT BASE é 768 para cada token de entrada, que também pode ser visto como uma representação vetorial real daquele determinado token."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "i6hRrVVjAiEB"
      },
      "source": [
        "## Pré Treinamento"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "goMCRE6XAkG2"
      },
      "source": [
        "O BERT possui 2 estratégias de pré treinamento: Masked Language Model (MLM) e Next Sentence Prediction (NSP)."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "wti5YhcECm_R"
      },
      "source": [
        "### **Masked LM (MLM)**"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "p2y55E72CqFh"
      },
      "source": [
        "Nessa abordagem, são mascarados ([MASK]) aleatoriamente $15\\%$ dos tokens de entrada e então feita a predição apenas desses tokens mascarados por meio de uma função $\\textit{softmax}$. Dentro dos $15\\%$ dos tokens aleatoriamente selecionados: \n",
        "\n",
        "\\\\\n",
        "\n",
        "* $80\\%$ das vezes é sustituído com [MASK]:\n",
        "\n",
        "$\\quad\\quad\\quad$ Fui ao banco $\\rightarrow$ Fui ao [MASK]\n",
        "\n",
        "\\\\\n",
        "\n",
        "* $10\\%$ das vezes é susbtituído com alguma palavra aleatória:\n",
        "\n",
        "$\\quad\\quad\\quad$ Fui ao banco $\\rightarrow$ Fui ao chovendo\n",
        "\n",
        "\\\\\n",
        "\n",
        "* $10\\%$ das vezes é mantido a mesma palavra\n",
        "\n",
        "$\\quad\\quad\\quad$ Fui ao banco $\\rightarrow$ Fui ao banco"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "xNizildmPYuA"
      },
      "source": [
        "Exemplo:\n",
        "\n",
        "Sequência de entrada: O homem [MASK] a loja com [MASK] cachorro\n",
        "\n",
        "Sequência alvo:                      **foi**     e          **seu**"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "UiST_osLMJLN"
      },
      "source": [
        "### **Next Sentence Prediction (NSP)**"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DIvyDwkXMO0a"
      },
      "source": [
        "Essa estratégia, por sua vez, se caracteriza por focar nas relações entre as sentenças, pré-treinando assim para uma tarefa de predição de próximas sentenças em que são possíveis apenas 2 resultados, \"IsNext\", caso a sentença B seja a próxima sentença de A ou \"NotNext\" caso contrário.\n",
        "\n",
        "Para essa tarefa, os dados de treino consistem de $50\\%$ dos dados rotulados como \"IsNext\" e os outros $50\\%$ como \"NotNext\", e é aplicada em conjunto com a estratégia de Masked ML explicada acima."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "xUfWL1aoJaBd"
      },
      "source": [
        "Exemplos:\n",
        "\n",
        "Input = [CLS] o homem [MASK] a loja [SEP]\n",
        "\n",
        "ele comprou uma garrafa de [MASK] leite [SEP]\n",
        "\n",
        "Label = IsNext\n",
        "\n",
        "\\\n",
        "\n",
        "Input = [CLS] o homem [MASK] a loja [SEP]\n",
        "\n",
        "o [MASK] está muito ensolarado [SEP]\n",
        "\n",
        "Label = NotNext\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2vDvFQfyRO1U"
      },
      "source": [
        "# Utilizando o BERT: **Fine-Tuning** ou **Feature Extraction**"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "TcOzTxZvRY1r"
      },
      "source": [
        "O modelo BERT pode ser utilizado por meio da **Extração de Features, técnica que foi utilizada nesse notebook** ou por meio da técnica chamada Fine-Tuning.\n",
        "\n",
        "Na **Extração de Features**, o modelo pré-treinado do Bert é usado para extrair features da base de dados em questão, de tal forma que essas possam ser usadas no treinamento de modelos com regressão Logística, Random Forest entre outros.\n",
        "\n",
        "Com o **Fine-Tuning**, é possível pegarmos o nosso modelo pré treinado e usar uma outra base de dados deseja para treinarmos mais parâmetros baseados baseados na tarefa em que desejamos executar, seja ela classificação de textos ou responder perguntas por exemplo. No Fine-Tuning,são usados modelos contidos no próprio pacote do BERT para execução de tarefas, diferentemente da extração de features, em que usamos modelos externos ao desse pacote.Você pode procurar mais a respeito nos seguintes links:\n",
        "\n",
        "(Classificação de Pares de Sentenças) [[6]](https://colab.research.google.com/drive/1pTuQhug6Dhl9XalKB0zUGf4FIdYFlpcX#scrollTo=9cQNvaZ9bnyy): É semelhante ao processo de Next Sentence Prediction (NSP), adicionando uma camada de classificação no topo do output referente ao token [CLS] retornando uma distribuição de probabilidades calculadas por uma função $\\textit{softmax}$  \n",
        "\n",
        "Tarefa de responder perguntas: Essa é uma tarefa de predição [[7]](https://mccormickml.com/2020/03/10/question-answering-with-a-fine-tuned-BERT/). O modelo recebe uma pergunta, que atua como a primeira sentença e um parágrafo contendo o contexto relacionado a pergunta, atuando como a segunda sentença. É feito então o produto escalar da forma final do embedding de cada token com um vetor de pesos, sendo aplicado em uma função de ativação para retornar uma distribuição de probabilidade."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FGjPbzinPxab"
      },
      "source": [
        "#Mão na Massa"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "YKswmOITHI9L"
      },
      "source": [
        "## Carregando os Dados"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "l8jG9pqj3thm"
      },
      "source": [
        "Instalando as bibliotecas necessárias:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "H20N1cso31zC"
      },
      "source": [
        "!pip install unidecode\n",
        "!pip install ftfy\n",
        "!pip install transformers==4.2.2\n",
        "!pip install pyreadr\n",
        "!pip install git+https://github.com/felipemaiapolo/legalnlp\n",
        "clear_output()"
      ],
      "execution_count": 9,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "OL5-IFzvxHhL"
      },
      "source": [
        "Aqui vamos importar algumas bibliotecas que serão muito úteis para frente!"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Xq6V2tDZI0x8"
      },
      "source": [
        "# manipulação numérica e de dataframes\n",
        "import numpy as np\n",
        "import pandas as pd\n",
        "\n",
        "# gráficos e ajustes visuais\n",
        "import seaborn as sns\n",
        "import matplotlib.pyplot as plt\n",
        "import textwrap\n",
        "from tqdm import tqdm\n",
        "\n",
        "# pré-processamento de textos e lidar com variáveis categóricas, modelos e métricas \n",
        "from sklearn.model_selection import train_test_split\n",
        "from sklearn.preprocessing import LabelEncoder\n",
        "\n",
        "# tratamento de textos \n",
        "import unidecode\n",
        "import re\n",
        "import ftfy"
      ],
      "execution_count": 10,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "rNfZlG8CJE5b"
      },
      "source": [
        "# Code for get data from github\n",
        "data=pd.read_csv('https://raw.githubusercontent.com/felipemaiapolo/legalnlp/main/demo/data_base.csv')\n",
        "data.drop(columns=['Unnamed: 0'],inplace=True)\n",
        "data.dropna(inplace=True)"
      ],
      "execution_count": 11,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "QdpuvKK_QLtq"
      },
      "source": [
        "## Breve exploração na base de dados"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "W2xU7K-6xQIe"
      },
      "source": [
        "Antes de aplicar os modelos, vale a pena dar uma explorada nos dados que estamos trabalhando. Primeiramente, qual o tamanho desse conjunto de dados?"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "in9zK0pXxim2",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "86f83109-2e31-4dae-8e6c-b48bc1690af8"
      },
      "source": [
        "print(\"Quantidade de linhas: \", data.shape[0])\n",
        "print(\"Quantidade de colunas: \", data.shape[1])"
      ],
      "execution_count": 12,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Quantidade de linhas:  6449\n",
            "Quantidade de colunas:  2\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "9Ft1fRf9x2qN"
      },
      "source": [
        "Vamos visualizar uma pequena amostra desses dados:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EHG1hRhow4_F",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 353
        },
        "outputId": "df07d3ff-d3d8-4ed2-b0c1-5c0b57217e8d"
      },
      "source": [
        "# amostra de tamanho 5\n",
        "data.sample(5)"
      ],
      "execution_count": 13,
      "outputs": [
        {
          "output_type": "execute_result",
          "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>text</th>\n",
              "      <th>label</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>1726</th>\n",
              "      <td>Certidão de Publicação Expedida Relação :0325/...</td>\n",
              "      <td>H:Suspenso</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2963</th>\n",
              "      <td>Certidão de Objeto e Pé Expedida Certidão - Ob...</td>\n",
              "      <td>H:Ativo</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4761</th>\n",
              "      <td>Tipo do Movimento:Recebidos os autos</td>\n",
              "      <td>H:Ativo</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4264</th>\n",
              "      <td>Tipo do Movimento:Arquivamento Tipo de arquiva...</td>\n",
              "      <td>H:Arquivado</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>5154</th>\n",
              "      <td>Petição Juntada Juntada a petição diversa - Ti...</td>\n",
              "      <td>H:Ativo</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                                                   text        label\n",
              "1726  Certidão de Publicação Expedida Relação :0325/...   H:Suspenso\n",
              "2963  Certidão de Objeto e Pé Expedida Certidão - Ob...      H:Ativo\n",
              "4761              Tipo do Movimento:Recebidos os autos       H:Ativo\n",
              "4264  Tipo do Movimento:Arquivamento Tipo de arquiva...  H:Arquivado\n",
              "5154  Petição Juntada Juntada a petição diversa - Ti...      H:Ativo"
            ]
          },
          "metadata": {},
          "execution_count": 13
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "NpecX-pNyK5-"
      },
      "source": [
        "Vamos verificar se tem algum valor faltante nas colunas:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "oY9LxHbC7213",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "a407b3c2-cc56-4a14-da66-66703d9211d7"
      },
      "source": [
        "# valores nas colunas faltante\n",
        "print(f'Total de: {data[\"text\"].isna().sum()} textos vazios')\n",
        "print(f'Total de: {data[\"label\"].isna().sum()} classes vazias')"
      ],
      "execution_count": 14,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Total de: 0 textos vazios\n",
            "Total de: 0 classes vazias\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "1Ogl5M1DygS8"
      },
      "source": [
        "Aqui um gráfico de como os rótulos da coluna \"label\" estão distribuídos:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KNfwYV91JqWi",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 54
        },
        "outputId": "2eeebf07-2fdf-47be-f7c3-64d908d9e730"
      },
      "source": [
        "# countplot das labels\n",
        "sns.countplot(x = data['label'])\n",
        "\n",
        "plt.xticks(np.arange(0, 3, step = 1), ['Arquivado', 'Ativo', 'Não Ativo'])\n",
        "plt.xlabel('Status')\n",
        "plt.ylabel('Quantidade')\n",
        "plt.title('Status countplot')\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": 15,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "9g76QyAEzNj6"
      },
      "source": [
        "E em frequências relativas:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "j73ryhTCyk9p",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 160
        },
        "outputId": "bd06586e-8f63-4f25-9a79-1a3282f89c53"
      },
      "source": [
        "# Frequência das labels\n",
        "freq = pd.DataFrame(data['label'].value_counts()/len(data))\n",
        "freq"
      ],
      "execution_count": 16,
      "outputs": [
        {
          "output_type": "execute_result",
          "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>label</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>H:Arquivado</th>\n",
              "      <td>0.471391</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>H:Ativo</th>\n",
              "      <td>0.452318</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>H:Suspenso</th>\n",
              "      <td>0.076291</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                label\n",
              "H:Arquivado  0.471391\n",
              "H:Ativo      0.452318\n",
              "H:Suspenso   0.076291"
            ]
          },
          "metadata": {},
          "execution_count": 16
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2TbkubgI9Br6"
      },
      "source": [
        "##Limpando os textos"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7YzICwOuEC81"
      },
      "source": [
        "Aqui vamos aplicar a função abaixo (clean_bert), que recebe um texto como argumento e realiza a limpeza desse texto, cuidando de Unicodes ruins (caracteres estranhos) por meio do atributo fix_text da biblioteca [ftfy: fixes text for you](https://ftfy.readthedocs.io/en/latest/) e fazendo substituições de determinados caracteres para outros específicos."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "28yDf4zhFWId"
      },
      "source": [
        "Com a função clean_bert, aplicamos na coluna de textos do nosso conjunto de dados:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kQ5P4VeyCTw_",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 829
        },
        "outputId": "01a1ada8-08ea-4078-f6d2-f7f1e1e59a37"
      },
      "source": [
        "from legalnlp.clean_functions import clean_bert\n",
        "data['text'] = data['text'].apply(lambda x:clean_bert(x))\n",
        "data"
      ],
      "execution_count": 17,
      "outputs": [
        {
          "output_type": "execute_result",
          "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>text</th>\n",
              "      <th>label</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>Início da Execução Juntado 0008003-11.2018.8.2...</td>\n",
              "      <td>H:Suspenso</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>Ofício Requisitório - Comunicação - Protocolo ...</td>\n",
              "      <td>H:Arquivado</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>Arquivado Provisoriamente aguardando manifestação</td>\n",
              "      <td>H:Suspenso</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>Petição Juntada Juntada a petição diversa - Ti...</td>\n",
              "      <td>H:Ativo</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>Remetidos os Autos para o Arquivo Geral – Devo...</td>\n",
              "      <td>H:Arquivado</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>6444</th>\n",
              "      <td>Remetidos os Autos para o Arquivo Geral – Devo...</td>\n",
              "      <td>H:Suspenso</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>6445</th>\n",
              "      <td>Certidão de Trânsito em Julgado com Baixa Expe...</td>\n",
              "      <td>H:Arquivado</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>6446</th>\n",
              "      <td>Petição Juntada Nº Protocolo: WSCB.19.70085805...</td>\n",
              "      <td>H:Ativo</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>6447</th>\n",
              "      <td>Remetidos os Autos para o Tribunal de Justiça/...</td>\n",
              "      <td>H:Ativo</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>6448</th>\n",
              "      <td>Arquivado Definitivamente</td>\n",
              "      <td>H:Arquivado</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>6449 rows × 2 columns</p>\n",
              "</div>"
            ],
            "text/plain": [
              "                                                   text        label\n",
              "0     Início da Execução Juntado 0008003-11.2018.8.2...   H:Suspenso\n",
              "1     Ofício Requisitório - Comunicação - Protocolo ...  H:Arquivado\n",
              "2     Arquivado Provisoriamente aguardando manifestação   H:Suspenso\n",
              "3     Petição Juntada Juntada a petição diversa - Ti...      H:Ativo\n",
              "4     Remetidos os Autos para o Arquivo Geral – Devo...  H:Arquivado\n",
              "...                                                 ...          ...\n",
              "6444  Remetidos os Autos para o Arquivo Geral – Devo...   H:Suspenso\n",
              "6445  Certidão de Trânsito em Julgado com Baixa Expe...  H:Arquivado\n",
              "6446  Petição Juntada Nº Protocolo: WSCB.19.70085805...      H:Ativo\n",
              "6447  Remetidos os Autos para o Tribunal de Justiça/...      H:Ativo\n",
              "6448                          Arquivado Definitivamente  H:Arquivado\n",
              "\n",
              "[6449 rows x 2 columns]"
            ]
          },
          "metadata": {},
          "execution_count": 17
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ygwPzZkb9jeJ"
      },
      "source": [
        "Após a limpeza dos dados, vamos dar uma olhada em alguns exemplos:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "mlmyxCMv9i_z",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "f8d5f4d1-a459-4ccb-ee6b-052606fe942f"
      },
      "source": [
        "str(data.loc[3, 'text'])"
      ],
      "execution_count": 18,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "application/vnd.google.colaboratory.intrinsic+json": {
              "type": "string"
            },
            "text/plain": [
              "'Petição Juntada Juntada a petição diversa - Tipo: Petição Intermediária em Procedimento Comum Cível - Número: 80031 - Protocolo: FGRU19000140154 - Complemento: autor, expedição de ofício.'"
            ]
          },
          "metadata": {},
          "execution_count": 18
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kT6TxRcg-OL2",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "5af35257-65dc-4bae-c828-5d9055b24d20"
      },
      "source": [
        "str(data.loc[1278, 'text'])"
      ],
      "execution_count": 19,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "application/vnd.google.colaboratory.intrinsic+json": {
              "type": "string"
            },
            "text/plain": [
              "'Certidão de Cartório Expedida Certifico e dou fé que foram interpostos 2 (dois) incidentes digitais para expedição de ofício requisitório sendo 1(um) referente a ORPV e 1 (um) referente a Precatório, os quais encontram-se aguardando análise. Nada Mais.'"
            ]
          },
          "metadata": {},
          "execution_count": 19
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lRmFU3wmE36f"
      },
      "source": [
        "#Aplicando o Label Enconder para deixar o target com valores númericos"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lbpDsfiZGGGg"
      },
      "source": [
        "Agora, com os textos limpos, vamos ver brevemente como lidar com as variáveis categóricas através do LabelEnconder: "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "d9K_PKu4vNDP",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 433
        },
        "outputId": "554f6ccf-f4d7-49a6-c9ce-e7a9cfefc123"
      },
      "source": [
        "encoder = LabelEncoder()\n",
        "encoder.fit(data['label'])\n",
        "data['encoded'] = encoder.transform(data['label'])\n",
        "data"
      ],
      "execution_count": 20,
      "outputs": [
        {
          "output_type": "execute_result",
          "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>text</th>\n",
              "      <th>label</th>\n",
              "      <th>encoded</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>Início da Execução Juntado 0008003-11.2018.8.2...</td>\n",
              "      <td>H:Suspenso</td>\n",
              "      <td>2</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>Ofício Requisitório - Comunicação - Protocolo ...</td>\n",
              "      <td>H:Arquivado</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>Arquivado Provisoriamente aguardando manifestação</td>\n",
              "      <td>H:Suspenso</td>\n",
              "      <td>2</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>Petição Juntada Juntada a petição diversa - Ti...</td>\n",
              "      <td>H:Ativo</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>Remetidos os Autos para o Arquivo Geral – Devo...</td>\n",
              "      <td>H:Arquivado</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>6444</th>\n",
              "      <td>Remetidos os Autos para o Arquivo Geral – Devo...</td>\n",
              "      <td>H:Suspenso</td>\n",
              "      <td>2</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>6445</th>\n",
              "      <td>Certidão de Trânsito em Julgado com Baixa Expe...</td>\n",
              "      <td>H:Arquivado</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>6446</th>\n",
              "      <td>Petição Juntada Nº Protocolo: WSCB.19.70085805...</td>\n",
              "      <td>H:Ativo</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>6447</th>\n",
              "      <td>Remetidos os Autos para o Tribunal de Justiça/...</td>\n",
              "      <td>H:Ativo</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>6448</th>\n",
              "      <td>Arquivado Definitivamente</td>\n",
              "      <td>H:Arquivado</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>6449 rows × 3 columns</p>\n",
              "</div>"
            ],
            "text/plain": [
              "                                                   text        label  encoded\n",
              "0     Início da Execução Juntado 0008003-11.2018.8.2...   H:Suspenso        2\n",
              "1     Ofício Requisitório - Comunicação - Protocolo ...  H:Arquivado        0\n",
              "2     Arquivado Provisoriamente aguardando manifestação   H:Suspenso        2\n",
              "3     Petição Juntada Juntada a petição diversa - Ti...      H:Ativo        1\n",
              "4     Remetidos os Autos para o Arquivo Geral – Devo...  H:Arquivado        0\n",
              "...                                                 ...          ...      ...\n",
              "6444  Remetidos os Autos para o Arquivo Geral – Devo...   H:Suspenso        2\n",
              "6445  Certidão de Trânsito em Julgado com Baixa Expe...  H:Arquivado        0\n",
              "6446  Petição Juntada Nº Protocolo: WSCB.19.70085805...      H:Ativo        1\n",
              "6447  Remetidos os Autos para o Tribunal de Justiça/...      H:Ativo        1\n",
              "6448                          Arquivado Definitivamente  H:Arquivado        0\n",
              "\n",
              "[6449 rows x 3 columns]"
            ]
          },
          "metadata": {},
          "execution_count": 20
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "NuKS1fcsN_yM"
      },
      "source": [
        "A coluna 'encoded' representa a forma que os elementos da coluna 'label' foram transformados em variáveis numéricas, como temos 3 classes, cada uma recebeu um valor 0, 1 ou 2, como podemos ver abaixo:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KaFe4eU9_DLm",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 160
        },
        "outputId": "1c1196fa-ec55-4865-87de-e1460c1a3b46"
      },
      "source": [
        "data.loc[[0, 1, 3], ['label', 'encoded']]"
      ],
      "execution_count": 21,
      "outputs": [
        {
          "output_type": "execute_result",
          "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>label</th>\n",
              "      <th>encoded</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>H:Suspenso</td>\n",
              "      <td>2</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>H:Arquivado</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>H:Ativo</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "         label  encoded\n",
              "0   H:Suspenso        2\n",
              "1  H:Arquivado        0\n",
              "3      H:Ativo        1"
            ]
          },
          "metadata": {},
          "execution_count": 21
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "o-UscTSdwnl3"
      },
      "source": [
        "#Aplicando o tokenizador"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "rtoyztVNVojQ"
      },
      "source": [
        "O processo de tokenização é o processo de divisão do texto a fim de identificarmos as diferentes palavras das quais ele é formado. No caso do BERT, temos um tokenizador próprio que além de focar na identificação de cada palavra do textos, também há um processo de reconhecimento de palavras que não estão no nosso vocabulário propriamente dito, o que é feito por meio da divisão dessas palavras na tokenização, de tal forma que um palavra pode gerar mais de um token. Dessa maneira, o modelo pode aprender as partes dessa palavra e baseado no contexto em que ela está inserada e no contexto de outras palavras que possuam tokens em comum com ela é possível \"aprender seu significado\"."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Z9p1LsZnwqAl",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "786afc39-421d-45ba-b6fd-f5d498a3eae4"
      },
      "source": [
        "import pickle\n",
        "import os\n",
        "import copy \n",
        "\n",
        "import torch\n",
        "import torch.nn as nn\n",
        "import torch.utils.data as tdata\n",
        "import torch.optim as optim\n",
        "import transformers\n",
        "from transformers import AutoModel, AutoTokenizer, AutoConfig\n",
        "from transformers import BertForPreTraining, BertModel, BertTokenizer, BertForMaskedLM, BertForNextSentencePrediction, BertForQuestionAnswering\n",
        "\n",
        "torch.cuda.is_available()"
      ],
      "execution_count": 22,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "True"
            ]
          },
          "metadata": {},
          "execution_count": 22
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lMuVyksyzJ9Y"
      },
      "source": [
        "# verificando se tem cuda disponível\n",
        "os.environ['CUDA_LAUNCH_BLOCKING'] = \"1\"\n",
        "device = torch.device('cuda:0')"
      ],
      "execution_count": 23,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "BrzvttrDzSu7"
      },
      "source": [
        "## Carregando o Tokenizador e o modelo pré treinado"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "phDfntkR7gEB"
      },
      "source": [
        "Aqui vamos criar duas variáveis que servirão como o tokenizador e o modelo BERT que foi pré-treinado. "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EsNBUeTEQVoc",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "518f910e-ef07-4ca7-e05f-21b1b123004c"
      },
      "source": [
        "from legalnlp.get_premodel import *\n",
        "# Fazendo o download do modelo pre-treinado BERTikal e o seu tokenizador\n",
        "get_premodel('bert')"
      ],
      "execution_count": 24,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "True"
            ]
          },
          "metadata": {},
          "execution_count": 24
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "x7gyDFE_zVbc",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "a4fce593-f144-48a5-d9a5-4551fcb708b2"
      },
      "source": [
        "# Caso esteja usando o Google Colab, não esqueça de ligar a GPU para o ambiente de execução\n",
        "%%time\n",
        "\n",
        "bert_model =  BertModel.from_pretrained('/content/BERTikal/').to(device)\n",
        "bert_tokenizer = BertTokenizer.from_pretrained('/content/BERTikal/vocab.txt', do_lower_case=False)\n",
        "\n",
        "clear_output()"
      ],
      "execution_count": 25,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "CPU times: user 3.78 s, sys: 1.33 s, total: 5.11 s\n",
            "Wall time: 5.13 s\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "tLp13bvi7wUk"
      },
      "source": [
        "Aqui as configurações do modelo:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "NdYtbQAy6cSK",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "b49bfbbe-0883-4233-fc47-6666270919c0"
      },
      "source": [
        "# configuração do BERT\n",
        "bert_model.config"
      ],
      "execution_count": 26,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "BertConfig {\n",
              "  \"_name_or_path\": \"/content/BERTikal/\",\n",
              "  \"architectures\": [\n",
              "    \"BertForMaskedLM\"\n",
              "  ],\n",
              "  \"attention_probs_dropout_prob\": 0.1,\n",
              "  \"directionality\": \"bidi\",\n",
              "  \"gradient_checkpointing\": false,\n",
              "  \"hidden_act\": \"gelu\",\n",
              "  \"hidden_dropout_prob\": 0.1,\n",
              "  \"hidden_size\": 768,\n",
              "  \"initializer_range\": 0.02,\n",
              "  \"intermediate_size\": 3072,\n",
              "  \"layer_norm_eps\": 1e-12,\n",
              "  \"max_position_embeddings\": 512,\n",
              "  \"model_type\": \"bert\",\n",
              "  \"num_attention_heads\": 12,\n",
              "  \"num_hidden_layers\": 12,\n",
              "  \"output_past\": true,\n",
              "  \"pad_token_id\": 0,\n",
              "  \"pooler_fc_size\": 768,\n",
              "  \"pooler_num_attention_heads\": 12,\n",
              "  \"pooler_num_fc_layers\": 3,\n",
              "  \"pooler_size_per_head\": 128,\n",
              "  \"pooler_type\": \"first_token_transform\",\n",
              "  \"position_embedding_type\": \"absolute\",\n",
              "  \"transformers_version\": \"4.2.2\",\n",
              "  \"type_vocab_size\": 2,\n",
              "  \"use_cache\": true,\n",
              "  \"vocab_size\": 29794\n",
              "}"
            ]
          },
          "metadata": {},
          "execution_count": 26
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "EpqYLuVkV60y"
      },
      "source": [
        "# Usando o tokenizador"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "IRhLI6Em8Fnq"
      },
      "source": [
        "Nessa seção veremos como o tokenizador importado acima funciona com alguns exemplos, relacionando com a teoria apresentada no começo desse tutorial."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jLw1S2z6A_lz",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "b118260a-f2a0-4f17-bac1-d6fad314fcf1"
      },
      "source": [
        "# Exemplo 1\n",
        "print(bert_tokenizer.tokenize('Example to test the text tokenizer'))\n",
        "print()\n",
        "\n",
        "# Exemplo 2\n",
        "print(bert_tokenizer.encode('Another example with the tokenizer'))\n",
        "print(bert_tokenizer.decode(bert_tokenizer.encode('Another example with the tokenizer')))\n",
        "print()\n",
        "\n",
        "# Exemplo 3\n",
        "tokens = bert_tokenizer.tokenize(\"This is a sample text to test the tokenizer.\")\n",
        "print(bert_tokenizer.convert_tokens_to_ids(tokens))\n",
        "print(bert_tokenizer.decode(bert_tokenizer.convert_tokens_to_ids(tokens)))"
      ],
      "execution_count": 27,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "['Ex', '##amp', '##le', 'to', 'test', 'the', 'tex', '##t', 'to', '##ken', '##ize', '##r']\n",
            "\n",
            "[101, 7181, 5238, 11367, 3066, 10125, 1621, 374, 8110, 15134, 22282, 102]\n",
            "[CLS] Another example with the tokenizer [SEP]\n",
            "\n",
            "[16989, 847, 123, 5748, 3066, 4509, 22286, 374, 15606, 1621, 374, 8110, 15134, 22282, 119]\n",
            "This is a sample text to test the tokenizer.\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kfTYJVWM2oaS",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "6e55d5ee-022b-47ed-e4b5-1a700afe21af"
      },
      "source": [
        "bert_tokenizer.convert_tokens_to_ids(['[CLS]', '[SEP]'])"
      ],
      "execution_count": 28,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[101, 102]"
            ]
          },
          "metadata": {},
          "execution_count": 28
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "S7Ia478J2RPG",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "05b8820c-a6d0-461d-86cb-3718128e9f61"
      },
      "source": [
        "# Tamaho do vocabulário\n",
        "print(\"Vocab size: \", bert_tokenizer.vocab_size)"
      ],
      "execution_count": 29,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Vocab size:  29794\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vYOnrLcECltx",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "d02537aa-3e47-4b44-e0dd-aa32bdb35e60"
      },
      "source": [
        "bert = data['text'].apply(lambda x: bert_tokenizer.encode(x, add_special_tokens=True,max_length=512, truncation = True))\n",
        "print('Max sentence length: ', max([len(sen) for sen in bert]))"
      ],
      "execution_count": 30,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Max sentence length:  512\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8Ob6jeHKB1WM"
      },
      "source": [
        "# Transformando nossos dados em tensores"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ck21AQ8RGAZg",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "40d8e09f-76d3-4459-f2d6-d56bc98b9ed7"
      },
      "source": [
        "# Fazendo a padronização dos textos\n",
        "wrapper = textwrap.TextWrapper()\n",
        "data_text = list(data['text'])\n",
        "\n",
        "for text in range(len(data_text[:4])):\n",
        "  print(f'{wrapper.fill(data_text[text])}')\n",
        "  print()"
      ],
      "execution_count": 31,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Início da Execução Juntado 0008003-11.2018.8.26.0002 - Cumprimento de\n",
            "sentença\n",
            "\n",
            "Ofício Requisitório - Comunicação - Protocolo Eletrônico - DEPRE\n",
            "\n",
            "Arquivado Provisoriamente aguardando manifestação\n",
            "\n",
            "Petição Juntada Juntada a petição diversa - Tipo: Petição\n",
            "Intermediária em Procedimento Comum Cível - Número: 80031 - Protocolo:\n",
            "FGRU19000140154 - Complemento: autor, expedição de ofício.\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "fs26gMieGO-g",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "baf087d9-06c5-4b35-d23d-566e49f07714"
      },
      "source": [
        "# Aplicando o bert_tokenizer em nosso dataset com um comprimento máximo de 512 tokens\n",
        "encoded_inputs = bert_tokenizer(data_text, padding=True, truncation=True, max_length=512, return_tensors=\"pt\")\n",
        "\n",
        "\n",
        "#Agora temos nossos encoded_input em um dicionário com 3 chaves\n",
        "encoded_inputs.keys()"
      ],
      "execution_count": 32,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "dict_keys(['input_ids', 'token_type_ids', 'attention_mask'])"
            ]
          },
          "metadata": {},
          "execution_count": 32
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2_w_7Qj6nJSm",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "4ee210bd-509c-453e-aba7-dc67e5382d6a"
      },
      "source": [
        "#Visualiando o primeiro texto após a aplicação do tokenizador\n",
        "print(encoded_inputs['input_ids'][0])\n",
        "\n",
        "print()\n",
        "\n",
        "# Mostrando o mesmo texto decodificado \n",
        "print(wrapper.fill(bert_tokenizer.decode(encoded_inputs['input_ids'][0])))"
      ],
      "execution_count": 33,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "tensor([  101,  1308,   826,   180,  9354,   182, 12661,   243,  6324, 21702,\n",
            "        22335,   118,  1433,   119,  6437,   119,  1015,   119,  2633,   119,\n",
            "         6324, 22313,   118,  2887,   350,  3551,   125, 12385,   102,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
            "            0,     0])\n",
            "\n",
            "[CLS] Início da Execução Juntado 0008003 - 11. 2018. 8. 26. 0002 -\n",
            "Cumprimento de sentença [SEP] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]\n",
            "[PAD] [PAD] [PAD] [PAD]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "WB05mU98IRI8"
      },
      "source": [
        "# Enviando os tensores para para a GPU\n",
        "input_ids = encoded_inputs['input_ids'].to(device)"
      ],
      "execution_count": 34,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wsf4vruMIwPm",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "e08fdd11-7f5d-4530-86ab-fb4e468f4555"
      },
      "source": [
        "# Criando o nosso vetor de features \n",
        "features = []\n",
        "\n",
        "# Aplicando o modelo pré-treinado em cada frase e adicionando-o ao nosso vetor\n",
        "\n",
        "for i in tqdm(range(len(data_text))):\n",
        "\n",
        "    with torch.no_grad():\n",
        "    \n",
        "      last_hidden_states = bert_model(input_ids[i:(i+1)])[1].cpu().numpy().reshape(-1).tolist()\n",
        "\n",
        "    features.append(last_hidden_states)\n"
      ],
      "execution_count": 35,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "100%|██████████| 6449/6449 [09:34<00:00, 11.23it/s]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SKK-MN4PVYtr",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "08957dfd-7de1-466f-e33f-9308e28a402b"
      },
      "source": [
        "# Criando um numpy array com as features extraidas\n",
        "features = np.array(features)\n",
        "features[:2]"
      ],
      "execution_count": 36,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([[-0.24129799, -0.05777963,  0.00084872, ..., -0.01299   ,\n",
              "        -0.26171428, -0.09836803],\n",
              "       [-0.02064875,  0.0622899 , -0.03670149, ...,  0.19874208,\n",
              "        -0.32606599, -0.40796778]])"
            ]
          },
          "metadata": {},
          "execution_count": 36
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Syx5bus-V2OF",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "d0fb1492-599a-45b7-c642-5412b325678f"
      },
      "source": [
        "# Printando o numero de linhas e de colunas das features extraídas\n",
        "print('Número de linhas: ', features.shape[0])\n",
        "print('Número de colunas: ', features.shape[1])"
      ],
      "execution_count": 37,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Número de linhas:  6449\n",
            "Número de colunas:  768\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ha4UpsIvJFrp"
      },
      "source": [
        "#Dividindo os dados em dados de treino e dados de teste"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0Wsac6uRaBPt",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 221
        },
        "outputId": "f56f4100-48f7-4444-c238-b11337ff533a"
      },
      "source": [
        "data.head(5)"
      ],
      "execution_count": 38,
      "outputs": [
        {
          "output_type": "execute_result",
          "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>text</th>\n",
              "      <th>label</th>\n",
              "      <th>encoded</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>Início da Execução Juntado 0008003-11.2018.8.2...</td>\n",
              "      <td>H:Suspenso</td>\n",
              "      <td>2</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>Ofício Requisitório - Comunicação - Protocolo ...</td>\n",
              "      <td>H:Arquivado</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>Arquivado Provisoriamente aguardando manifestação</td>\n",
              "      <td>H:Suspenso</td>\n",
              "      <td>2</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>Petição Juntada Juntada a petição diversa - Ti...</td>\n",
              "      <td>H:Ativo</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>Remetidos os Autos para o Arquivo Geral – Devo...</td>\n",
              "      <td>H:Arquivado</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                                                text        label  encoded\n",
              "0  Início da Execução Juntado 0008003-11.2018.8.2...   H:Suspenso        2\n",
              "1  Ofício Requisitório - Comunicação - Protocolo ...  H:Arquivado        0\n",
              "2  Arquivado Provisoriamente aguardando manifestação   H:Suspenso        2\n",
              "3  Petição Juntada Juntada a petição diversa - Ti...      H:Ativo        1\n",
              "4  Remetidos os Autos para o Arquivo Geral – Devo...  H:Arquivado        0"
            ]
          },
          "metadata": {},
          "execution_count": 38
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Xv6dDOjgahCT",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "e806c213-ad21-454c-c8cc-c6b9add2befb"
      },
      "source": [
        "features"
      ],
      "execution_count": 39,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([[-0.24129799, -0.05777963,  0.00084872, ..., -0.01299   ,\n",
              "        -0.26171428, -0.09836803],\n",
              "       [-0.02064875,  0.0622899 , -0.03670149, ...,  0.19874208,\n",
              "        -0.32606599, -0.40796778],\n",
              "       [-0.17507128,  0.02626378, -0.06640682, ..., -0.08320417,\n",
              "        -0.21811467,  0.04786057],\n",
              "       ...,\n",
              "       [-0.20186585, -0.13697383,  0.20084573, ..., -0.17613822,\n",
              "        -0.01791709, -0.3479954 ],\n",
              "       [ 0.33393225, -0.08231198,  0.00543304, ...,  0.1858722 ,\n",
              "         0.10800666, -0.03800488],\n",
              "       [-0.13628896,  0.15502027, -0.04423159, ..., -0.22792308,\n",
              "        -0.34851524,  0.01096392]])"
            ]
          },
          "metadata": {},
          "execution_count": 39
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "t1gN2_4VZwuH",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "0b6dd7d7-ea50-4bbd-abd9-4584fd70704e"
      },
      "source": [
        "df_features = pd.DataFrame(features)\n",
        "features_label = pd.concat([df_features, data['encoded']], axis = 1)\n",
        "features_label.shape"
      ],
      "execution_count": 40,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(6449, 769)"
            ]
          },
          "metadata": {},
          "execution_count": 40
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "hy-nkVge5QTy",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 250
        },
        "outputId": "a92c54bf-b00f-4617-84ec-306761c182f4"
      },
      "source": [
        "features_label.head(5)"
      ],
      "execution_count": 41,
      "outputs": [
        {
          "output_type": "execute_result",
          "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>0</th>\n",
              "      <th>1</th>\n",
              "      <th>2</th>\n",
              "      <th>3</th>\n",
              "      <th>4</th>\n",
              "      <th>5</th>\n",
              "      <th>6</th>\n",
              "      <th>7</th>\n",
              "      <th>8</th>\n",
              "      <th>9</th>\n",
              "      <th>10</th>\n",
              "      <th>11</th>\n",
              "      <th>12</th>\n",
              "      <th>13</th>\n",
              "      <th>14</th>\n",
              "      <th>15</th>\n",
              "      <th>16</th>\n",
              "      <th>17</th>\n",
              "      <th>18</th>\n",
              "      <th>19</th>\n",
              "      <th>20</th>\n",
              "      <th>21</th>\n",
              "      <th>22</th>\n",
              "      <th>23</th>\n",
              "      <th>24</th>\n",
              "      <th>25</th>\n",
              "      <th>26</th>\n",
              "      <th>27</th>\n",
              "      <th>28</th>\n",
              "      <th>29</th>\n",
              "      <th>30</th>\n",
              "      <th>31</th>\n",
              "      <th>32</th>\n",
              "      <th>33</th>\n",
              "      <th>34</th>\n",
              "      <th>35</th>\n",
              "      <th>36</th>\n",
              "      <th>37</th>\n",
              "      <th>38</th>\n",
              "      <th>39</th>\n",
              "      <th>...</th>\n",
              "      <th>729</th>\n",
              "      <th>730</th>\n",
              "      <th>731</th>\n",
              "      <th>732</th>\n",
              "      <th>733</th>\n",
              "      <th>734</th>\n",
              "      <th>735</th>\n",
              "      <th>736</th>\n",
              "      <th>737</th>\n",
              "      <th>738</th>\n",
              "      <th>739</th>\n",
              "      <th>740</th>\n",
              "      <th>741</th>\n",
              "      <th>742</th>\n",
              "      <th>743</th>\n",
              "      <th>744</th>\n",
              "      <th>745</th>\n",
              "      <th>746</th>\n",
              "      <th>747</th>\n",
              "      <th>748</th>\n",
              "      <th>749</th>\n",
              "      <th>750</th>\n",
              "      <th>751</th>\n",
              "      <th>752</th>\n",
              "      <th>753</th>\n",
              "      <th>754</th>\n",
              "      <th>755</th>\n",
              "      <th>756</th>\n",
              "      <th>757</th>\n",
              "      <th>758</th>\n",
              "      <th>759</th>\n",
              "      <th>760</th>\n",
              "      <th>761</th>\n",
              "      <th>762</th>\n",
              "      <th>763</th>\n",
              "      <th>764</th>\n",
              "      <th>765</th>\n",
              "      <th>766</th>\n",
              "      <th>767</th>\n",
              "      <th>encoded</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>-0.241298</td>\n",
              "      <td>-0.057780</td>\n",
              "      <td>0.000849</td>\n",
              "      <td>-0.048564</td>\n",
              "      <td>-0.056002</td>\n",
              "      <td>0.175353</td>\n",
              "      <td>0.991059</td>\n",
              "      <td>-0.050905</td>\n",
              "      <td>0.316997</td>\n",
              "      <td>-0.151544</td>\n",
              "      <td>-0.896808</td>\n",
              "      <td>-0.218365</td>\n",
              "      <td>-0.057039</td>\n",
              "      <td>-0.053357</td>\n",
              "      <td>-0.243437</td>\n",
              "      <td>-0.060434</td>\n",
              "      <td>-0.074569</td>\n",
              "      <td>-0.075404</td>\n",
              "      <td>0.804686</td>\n",
              "      <td>0.888134</td>\n",
              "      <td>0.640193</td>\n",
              "      <td>0.172075</td>\n",
              "      <td>-0.224678</td>\n",
              "      <td>0.069350</td>\n",
              "      <td>0.112654</td>\n",
              "      <td>-0.115589</td>\n",
              "      <td>0.000246</td>\n",
              "      <td>-0.027467</td>\n",
              "      <td>0.004945</td>\n",
              "      <td>-0.089470</td>\n",
              "      <td>-0.308828</td>\n",
              "      <td>-0.909054</td>\n",
              "      <td>-0.821151</td>\n",
              "      <td>-0.119412</td>\n",
              "      <td>0.029579</td>\n",
              "      <td>-0.200819</td>\n",
              "      <td>0.122889</td>\n",
              "      <td>-0.116986</td>\n",
              "      <td>0.744767</td>\n",
              "      <td>0.252832</td>\n",
              "      <td>...</td>\n",
              "      <td>-0.028599</td>\n",
              "      <td>0.357689</td>\n",
              "      <td>0.028028</td>\n",
              "      <td>0.077504</td>\n",
              "      <td>0.161417</td>\n",
              "      <td>0.284518</td>\n",
              "      <td>0.048083</td>\n",
              "      <td>-0.084286</td>\n",
              "      <td>0.096878</td>\n",
              "      <td>-0.236369</td>\n",
              "      <td>-0.055770</td>\n",
              "      <td>-0.818162</td>\n",
              "      <td>0.031071</td>\n",
              "      <td>0.054099</td>\n",
              "      <td>0.221157</td>\n",
              "      <td>0.080103</td>\n",
              "      <td>0.239643</td>\n",
              "      <td>-0.052498</td>\n",
              "      <td>-0.067606</td>\n",
              "      <td>0.637069</td>\n",
              "      <td>0.221610</td>\n",
              "      <td>0.155294</td>\n",
              "      <td>-0.234776</td>\n",
              "      <td>-0.163705</td>\n",
              "      <td>-0.304676</td>\n",
              "      <td>-0.079111</td>\n",
              "      <td>-0.142781</td>\n",
              "      <td>-0.690279</td>\n",
              "      <td>0.052602</td>\n",
              "      <td>-0.131286</td>\n",
              "      <td>-0.094529</td>\n",
              "      <td>-0.086117</td>\n",
              "      <td>0.324010</td>\n",
              "      <td>0.031410</td>\n",
              "      <td>-0.148010</td>\n",
              "      <td>-0.161888</td>\n",
              "      <td>-0.012990</td>\n",
              "      <td>-0.261714</td>\n",
              "      <td>-0.098368</td>\n",
              "      <td>2</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>-0.020649</td>\n",
              "      <td>0.062290</td>\n",
              "      <td>-0.036701</td>\n",
              "      <td>-0.129206</td>\n",
              "      <td>-0.030147</td>\n",
              "      <td>0.267447</td>\n",
              "      <td>0.985803</td>\n",
              "      <td>0.404766</td>\n",
              "      <td>-0.043943</td>\n",
              "      <td>-0.008192</td>\n",
              "      <td>-0.965343</td>\n",
              "      <td>-0.036896</td>\n",
              "      <td>0.174238</td>\n",
              "      <td>-0.086197</td>\n",
              "      <td>-0.253857</td>\n",
              "      <td>-0.178252</td>\n",
              "      <td>-0.129426</td>\n",
              "      <td>0.088896</td>\n",
              "      <td>0.983750</td>\n",
              "      <td>0.514609</td>\n",
              "      <td>0.003525</td>\n",
              "      <td>0.316241</td>\n",
              "      <td>-0.273741</td>\n",
              "      <td>0.133143</td>\n",
              "      <td>-0.276910</td>\n",
              "      <td>0.338380</td>\n",
              "      <td>0.248213</td>\n",
              "      <td>-0.211505</td>\n",
              "      <td>0.065028</td>\n",
              "      <td>0.215943</td>\n",
              "      <td>-0.362236</td>\n",
              "      <td>-0.894944</td>\n",
              "      <td>-0.902823</td>\n",
              "      <td>-0.003573</td>\n",
              "      <td>0.280066</td>\n",
              "      <td>-0.342073</td>\n",
              "      <td>-0.212170</td>\n",
              "      <td>0.014028</td>\n",
              "      <td>0.922013</td>\n",
              "      <td>0.341220</td>\n",
              "      <td>...</td>\n",
              "      <td>-0.189641</td>\n",
              "      <td>0.348583</td>\n",
              "      <td>-0.146080</td>\n",
              "      <td>-0.178621</td>\n",
              "      <td>0.088064</td>\n",
              "      <td>-0.132068</td>\n",
              "      <td>-0.175349</td>\n",
              "      <td>0.073264</td>\n",
              "      <td>-0.192409</td>\n",
              "      <td>-0.300864</td>\n",
              "      <td>-0.139611</td>\n",
              "      <td>-0.987249</td>\n",
              "      <td>0.156078</td>\n",
              "      <td>0.037288</td>\n",
              "      <td>0.285448</td>\n",
              "      <td>0.115848</td>\n",
              "      <td>0.349242</td>\n",
              "      <td>-0.180449</td>\n",
              "      <td>0.178451</td>\n",
              "      <td>-0.258686</td>\n",
              "      <td>0.088996</td>\n",
              "      <td>0.491710</td>\n",
              "      <td>0.192816</td>\n",
              "      <td>-0.203070</td>\n",
              "      <td>-0.036393</td>\n",
              "      <td>0.169442</td>\n",
              "      <td>-0.012561</td>\n",
              "      <td>-0.784642</td>\n",
              "      <td>0.203821</td>\n",
              "      <td>-0.327721</td>\n",
              "      <td>0.221271</td>\n",
              "      <td>0.035829</td>\n",
              "      <td>0.120155</td>\n",
              "      <td>-0.377430</td>\n",
              "      <td>-0.018894</td>\n",
              "      <td>0.236955</td>\n",
              "      <td>0.198742</td>\n",
              "      <td>-0.326066</td>\n",
              "      <td>-0.407968</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>-0.175071</td>\n",
              "      <td>0.026264</td>\n",
              "      <td>-0.066407</td>\n",
              "      <td>-0.049350</td>\n",
              "      <td>-0.165291</td>\n",
              "      <td>0.207093</td>\n",
              "      <td>0.989866</td>\n",
              "      <td>-0.081967</td>\n",
              "      <td>-0.088869</td>\n",
              "      <td>0.128043</td>\n",
              "      <td>-0.961851</td>\n",
              "      <td>-0.334999</td>\n",
              "      <td>-0.007591</td>\n",
              "      <td>0.016222</td>\n",
              "      <td>0.329067</td>\n",
              "      <td>-0.086552</td>\n",
              "      <td>-0.042560</td>\n",
              "      <td>0.146579</td>\n",
              "      <td>0.931600</td>\n",
              "      <td>0.756115</td>\n",
              "      <td>0.273291</td>\n",
              "      <td>0.242544</td>\n",
              "      <td>-0.298596</td>\n",
              "      <td>0.200004</td>\n",
              "      <td>0.284772</td>\n",
              "      <td>-0.262865</td>\n",
              "      <td>0.121457</td>\n",
              "      <td>-0.141515</td>\n",
              "      <td>0.085980</td>\n",
              "      <td>-0.101485</td>\n",
              "      <td>-0.125829</td>\n",
              "      <td>-0.920414</td>\n",
              "      <td>-0.797177</td>\n",
              "      <td>0.059803</td>\n",
              "      <td>-0.284483</td>\n",
              "      <td>-0.063034</td>\n",
              "      <td>0.230420</td>\n",
              "      <td>0.118417</td>\n",
              "      <td>0.768116</td>\n",
              "      <td>0.268125</td>\n",
              "      <td>...</td>\n",
              "      <td>0.144556</td>\n",
              "      <td>0.236306</td>\n",
              "      <td>-0.078471</td>\n",
              "      <td>0.180531</td>\n",
              "      <td>0.102452</td>\n",
              "      <td>0.103105</td>\n",
              "      <td>0.177969</td>\n",
              "      <td>-0.229650</td>\n",
              "      <td>-0.066170</td>\n",
              "      <td>-0.173536</td>\n",
              "      <td>-0.048938</td>\n",
              "      <td>-0.942150</td>\n",
              "      <td>0.008241</td>\n",
              "      <td>0.147699</td>\n",
              "      <td>-0.031388</td>\n",
              "      <td>0.023140</td>\n",
              "      <td>0.013995</td>\n",
              "      <td>-0.229492</td>\n",
              "      <td>0.137982</td>\n",
              "      <td>0.532035</td>\n",
              "      <td>0.349577</td>\n",
              "      <td>0.091968</td>\n",
              "      <td>-0.011063</td>\n",
              "      <td>-0.137871</td>\n",
              "      <td>0.297167</td>\n",
              "      <td>0.001698</td>\n",
              "      <td>-0.102495</td>\n",
              "      <td>-0.728532</td>\n",
              "      <td>-0.233642</td>\n",
              "      <td>0.036566</td>\n",
              "      <td>-0.098382</td>\n",
              "      <td>-0.247678</td>\n",
              "      <td>0.217619</td>\n",
              "      <td>0.058316</td>\n",
              "      <td>-0.155054</td>\n",
              "      <td>-0.036148</td>\n",
              "      <td>-0.083204</td>\n",
              "      <td>-0.218115</td>\n",
              "      <td>0.047861</td>\n",
              "      <td>2</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>0.014324</td>\n",
              "      <td>-0.329025</td>\n",
              "      <td>0.024853</td>\n",
              "      <td>0.222323</td>\n",
              "      <td>-0.014267</td>\n",
              "      <td>0.257152</td>\n",
              "      <td>0.991074</td>\n",
              "      <td>0.048891</td>\n",
              "      <td>0.241455</td>\n",
              "      <td>-0.222435</td>\n",
              "      <td>-0.905999</td>\n",
              "      <td>0.071710</td>\n",
              "      <td>0.111401</td>\n",
              "      <td>0.010447</td>\n",
              "      <td>-0.184156</td>\n",
              "      <td>-0.111745</td>\n",
              "      <td>-0.189662</td>\n",
              "      <td>0.136783</td>\n",
              "      <td>0.939135</td>\n",
              "      <td>0.883227</td>\n",
              "      <td>0.366644</td>\n",
              "      <td>0.002272</td>\n",
              "      <td>0.045706</td>\n",
              "      <td>-0.058188</td>\n",
              "      <td>0.097100</td>\n",
              "      <td>-0.241113</td>\n",
              "      <td>0.034688</td>\n",
              "      <td>-0.065021</td>\n",
              "      <td>-0.073843</td>\n",
              "      <td>-0.268549</td>\n",
              "      <td>-0.067699</td>\n",
              "      <td>-0.940219</td>\n",
              "      <td>-0.920578</td>\n",
              "      <td>-0.098243</td>\n",
              "      <td>0.075057</td>\n",
              "      <td>-0.084626</td>\n",
              "      <td>-0.105071</td>\n",
              "      <td>-0.321198</td>\n",
              "      <td>0.914201</td>\n",
              "      <td>0.089570</td>\n",
              "      <td>...</td>\n",
              "      <td>0.039434</td>\n",
              "      <td>0.178928</td>\n",
              "      <td>0.187277</td>\n",
              "      <td>0.101045</td>\n",
              "      <td>-0.182704</td>\n",
              "      <td>-0.142024</td>\n",
              "      <td>-0.020638</td>\n",
              "      <td>-0.023329</td>\n",
              "      <td>-0.099853</td>\n",
              "      <td>0.136789</td>\n",
              "      <td>-0.096149</td>\n",
              "      <td>-0.899521</td>\n",
              "      <td>0.157630</td>\n",
              "      <td>-0.100842</td>\n",
              "      <td>0.020823</td>\n",
              "      <td>-0.069071</td>\n",
              "      <td>0.326456</td>\n",
              "      <td>0.201793</td>\n",
              "      <td>-0.010377</td>\n",
              "      <td>0.471601</td>\n",
              "      <td>0.084335</td>\n",
              "      <td>-0.063110</td>\n",
              "      <td>-0.124073</td>\n",
              "      <td>0.037644</td>\n",
              "      <td>-0.377467</td>\n",
              "      <td>0.111129</td>\n",
              "      <td>-0.018057</td>\n",
              "      <td>-0.798761</td>\n",
              "      <td>0.006616</td>\n",
              "      <td>-0.180005</td>\n",
              "      <td>-0.106930</td>\n",
              "      <td>-0.078986</td>\n",
              "      <td>0.216194</td>\n",
              "      <td>-0.059559</td>\n",
              "      <td>-0.114143</td>\n",
              "      <td>0.021033</td>\n",
              "      <td>0.214921</td>\n",
              "      <td>-0.033294</td>\n",
              "      <td>-0.224924</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>0.230466</td>\n",
              "      <td>-0.112876</td>\n",
              "      <td>0.026541</td>\n",
              "      <td>-0.216750</td>\n",
              "      <td>-0.287071</td>\n",
              "      <td>0.177000</td>\n",
              "      <td>0.997694</td>\n",
              "      <td>-0.065976</td>\n",
              "      <td>0.089110</td>\n",
              "      <td>0.299920</td>\n",
              "      <td>-0.989716</td>\n",
              "      <td>-0.255239</td>\n",
              "      <td>0.163111</td>\n",
              "      <td>0.007347</td>\n",
              "      <td>0.231630</td>\n",
              "      <td>-0.029464</td>\n",
              "      <td>-0.053921</td>\n",
              "      <td>0.285242</td>\n",
              "      <td>0.985457</td>\n",
              "      <td>0.715217</td>\n",
              "      <td>0.335801</td>\n",
              "      <td>0.346124</td>\n",
              "      <td>-0.271849</td>\n",
              "      <td>0.415113</td>\n",
              "      <td>0.221321</td>\n",
              "      <td>0.119555</td>\n",
              "      <td>0.122984</td>\n",
              "      <td>-0.149265</td>\n",
              "      <td>0.001878</td>\n",
              "      <td>-0.111310</td>\n",
              "      <td>-0.241509</td>\n",
              "      <td>-0.948832</td>\n",
              "      <td>-0.934402</td>\n",
              "      <td>-0.136327</td>\n",
              "      <td>0.002776</td>\n",
              "      <td>-0.318112</td>\n",
              "      <td>0.149392</td>\n",
              "      <td>-0.060772</td>\n",
              "      <td>0.955385</td>\n",
              "      <td>0.181874</td>\n",
              "      <td>...</td>\n",
              "      <td>-0.148026</td>\n",
              "      <td>0.064792</td>\n",
              "      <td>0.033269</td>\n",
              "      <td>-0.175626</td>\n",
              "      <td>0.051939</td>\n",
              "      <td>0.352942</td>\n",
              "      <td>0.010405</td>\n",
              "      <td>-0.357216</td>\n",
              "      <td>-0.252298</td>\n",
              "      <td>-0.152176</td>\n",
              "      <td>-0.014635</td>\n",
              "      <td>-0.979448</td>\n",
              "      <td>0.218483</td>\n",
              "      <td>0.111220</td>\n",
              "      <td>0.271564</td>\n",
              "      <td>0.314034</td>\n",
              "      <td>0.015955</td>\n",
              "      <td>-0.274112</td>\n",
              "      <td>0.093881</td>\n",
              "      <td>0.391308</td>\n",
              "      <td>0.389126</td>\n",
              "      <td>0.006711</td>\n",
              "      <td>-0.194761</td>\n",
              "      <td>-0.392457</td>\n",
              "      <td>0.360414</td>\n",
              "      <td>-0.172945</td>\n",
              "      <td>-0.323331</td>\n",
              "      <td>-0.906382</td>\n",
              "      <td>0.169631</td>\n",
              "      <td>0.062481</td>\n",
              "      <td>-0.057736</td>\n",
              "      <td>0.249552</td>\n",
              "      <td>0.596583</td>\n",
              "      <td>0.229108</td>\n",
              "      <td>-0.349671</td>\n",
              "      <td>0.306246</td>\n",
              "      <td>0.112156</td>\n",
              "      <td>-0.077146</td>\n",
              "      <td>-0.206237</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>5 rows × 769 columns</p>\n",
              "</div>"
            ],
            "text/plain": [
              "          0         1         2  ...       766       767  encoded\n",
              "0 -0.241298 -0.057780  0.000849  ... -0.261714 -0.098368        2\n",
              "1 -0.020649  0.062290 -0.036701  ... -0.326066 -0.407968        0\n",
              "2 -0.175071  0.026264 -0.066407  ... -0.218115  0.047861        2\n",
              "3  0.014324 -0.329025  0.024853  ... -0.033294 -0.224924        1\n",
              "4  0.230466 -0.112876  0.026541  ... -0.077146 -0.206237        0\n",
              "\n",
              "[5 rows x 769 columns]"
            ]
          },
          "metadata": {},
          "execution_count": 41
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_d-eEWJc2fbe"
      },
      "source": [
        "X_train, X_test, y_train, y_test = train_test_split(features_label.drop(columns = ['encoded']), features_label['encoded'], random_state = random_seed,test_size = 0.3)"
      ],
      "execution_count": 42,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "fU7Aepou-cWS",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "df5c8b75-370c-4490-8249-ae7362ee9c32"
      },
      "source": [
        "# Tamanhos dos x e y de treino e teste\n",
        "print(X_train.shape)\n",
        "print(X_test.shape)\n",
        "print(y_train.shape)\n",
        "print(y_test.shape)"
      ],
      "execution_count": 43,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "(4514, 768)\n",
            "(1935, 768)\n",
            "(4514,)\n",
            "(1935,)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "arI885saAgFk"
      },
      "source": [
        "#Classificação"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ISBqUzBaSQpW"
      },
      "source": [
        "from sklearn.metrics import accuracy_score, confusion_matrix, classification_report"
      ],
      "execution_count": 44,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UbC0euB0d-at"
      },
      "source": [
        "def bootstrap(X_test, y_test, model, nn=False, B=250):\n",
        "    #Criando dicinário para armazenar os resultados\n",
        "    out={}\n",
        "    out['accuracy']=[]\n",
        "    out['macro avg']={}\n",
        "    out['macro avg']['f1-score']=[]\n",
        "    out['macro avg']['recall']=[]\n",
        "    out['macro avg']['precision']=[]\n",
        "    out['weighted avg']={}\n",
        "    out['weighted avg']['f1-score']=[]\n",
        "    out['weighted avg']['recall']=[]\n",
        "    out['weighted avg']['precision']=[]\n",
        "\n",
        "    #Aplicando Bootstrap no conjunto de teste\n",
        "    for b in tqdm(range(B)):\n",
        "        ind = np.random.choice(range(y_test.shape[0]),y_test.shape[0])\n",
        "        X_test_boot, y_test_boot = X_test[ind,:], y_test[ind]\n",
        "\n",
        "        y_pred=model.predict(X_test_boot)\n",
        "        \n",
        "        if nn:\n",
        "            y_pred=np.argmax(y_pred,axis=1)\n",
        "            report=classification_report(y_test_boot, y_pred, labels=[0, 1, 2], output_dict=True)\n",
        "        else:\n",
        "            report=classification_report(y_test_boot, y_pred, labels=[0, 1, 2], output_dict=True)\n",
        "\n",
        "        out['accuracy'].append(report['accuracy'])\n",
        "        out['macro avg']['f1-score'].append(report['macro avg']['f1-score'])\n",
        "        out['macro avg']['recall'].append(report['macro avg']['recall'])\n",
        "        out['macro avg']['precision'].append(report['macro avg']['precision'])\n",
        "        out['weighted avg']['f1-score'].append(report['weighted avg']['f1-score'])\n",
        "        out['weighted avg']['recall'].append(report['weighted avg']['recall'])\n",
        "        out['weighted avg']['precision'].append(report['weighted avg']['precision'])\n",
        "\n",
        "    #Preparando a saída\n",
        "    y_pred=model.predict(X_test)\n",
        "    \n",
        "    if nn:\n",
        "        y_pred=np.argmax(y_pred,axis=1)\n",
        "        report=classification_report(y_test, y_pred, labels=[0, 1, 2], output_dict=True)\n",
        "    else:\n",
        "        report=classification_report(y_test, y_pred, labels=[0, 1, 2], output_dict=True)\n",
        "\n",
        "    out['accuracy'] = [report['accuracy'], np.std(out['accuracy'])]\n",
        "    out['macro avg']['f1-score'] = [report['macro avg']['f1-score'], np.std(out['macro avg']['f1-score'])] \n",
        "    out['macro avg']['recall'] = [report['macro avg']['recall'], np.std(out['macro avg']['recall'])] \n",
        "    out['macro avg']['precision'] = [report['macro avg']['precision'], np.std(out['macro avg']['precision'])] \n",
        "    out['weighted avg']['f1-score'] = [report['weighted avg']['f1-score'], np.std(out['weighted avg']['f1-score'])] \n",
        "    out['weighted avg']['recall'] = [report['weighted avg']['recall'], np.std(out['weighted avg']['recall'])] \n",
        "    out['weighted avg']['precision'] = [report['weighted avg']['precision'], np.std(out['weighted avg']['precision'])]\n",
        "    \n",
        "    return out"
      ],
      "execution_count": 45,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "KsZQ6vAo9zNa"
      },
      "source": [
        "## O Modelo de Boosting: CatBoost"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "1wXF1SAB0_41"
      },
      "source": [
        "Nesa seção vamos aplicar um modelo chamado CatBoost (Categorical Boosting), que é um modelo de aprendizado por comitê (ensemble learning)."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "hrDkioOPwP2b"
      },
      "source": [
        "!pip install catboost\n",
        "clear_output()"
      ],
      "execution_count": 46,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "rFuGg3Fl_TtB"
      },
      "source": [
        "# Importando o CatBoostClassifier \n",
        "from catboost import CatBoostClassifier"
      ],
      "execution_count": 47,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9jbhEnrnNY-u"
      },
      "source": [
        "# criando conjuntos de validação\n",
        "X_train, X_val, y_train, y_val = train_test_split(X_train, y_train, test_size = 0.1, stratify = y_train, random_state = random_seed)"
      ],
      "execution_count": 48,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "nYVahG36d4Y-",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "8514e687-0f47-4bb1-a09d-d83ef20db666"
      },
      "source": [
        "tunned_model = CatBoostClassifier(\n",
        "    loss_function = 'MultiClass',\n",
        "    random_seed=random_seed,\n",
        ")\n",
        "\n",
        "tunned_model.fit(\n",
        "    X_train, y_train,\n",
        "    verbose=500,\n",
        "    eval_set=(X_val, y_val),\n",
        "    early_stopping_rounds = 100\n",
        ")"
      ],
      "execution_count": 49,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Learning rate set to 0.111046\n",
            "0:\tlearn: 0.9914584\ttest: 0.9878537\tbest: 0.9878537 (0)\ttotal: 805ms\tremaining: 13m 23s\n",
            "Stopped by overfitting detector  (100 iterations wait)\n",
            "\n",
            "bestTest = 0.3548586212\n",
            "bestIteration = 259\n",
            "\n",
            "Shrink model to first 260 iterations.\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<catboost.core.CatBoostClassifier at 0x7f87d7d8b450>"
            ]
          },
          "metadata": {},
          "execution_count": 49
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "k5dWYb8bhr4j"
      },
      "source": [
        "y_cat_pred = tunned_model.predict(X_test)"
      ],
      "execution_count": 50,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "WkVrr31jhyOb",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "71715eb5-f823-49fa-bca4-68cd9b72759f"
      },
      "source": [
        "accuracy_score(y_test, y_cat_pred)"
      ],
      "execution_count": 51,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0.8594315245478036"
            ]
          },
          "metadata": {},
          "execution_count": 51
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DO-ficbleLd7",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "e44b9e6f-07f0-430e-f07c-d2983c5cc70e"
      },
      "source": [
        "tunned_model.predict_proba(X_test)[:5]"
      ],
      "execution_count": 52,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([[0.35976293, 0.56639144, 0.07384564],\n",
              "       [0.25827792, 0.66367198, 0.0780501 ],\n",
              "       [0.18339311, 0.79520801, 0.02139889],\n",
              "       [0.99052015, 0.00443338, 0.00504647],\n",
              "       [0.0833798 , 0.85763858, 0.05898163]])"
            ]
          },
          "metadata": {},
          "execution_count": 52
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jekdhPrrtSYk",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 301
        },
        "outputId": "1d77df4a-3282-4f1e-c424-dad356d69edc"
      },
      "source": [
        "cm = confusion_matrix(y_test, y_cat_pred)\n",
        "sns.heatmap(cm, annot = True, cmap = 'Greens', fmt = '.3g')\n",
        "plt.title(\"Heatmap\")\n",
        "plt.show()"
      ],
      "execution_count": 53,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FLKu3kC0tSYt",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "106fdb8d-c438-4e95-d9ba-fd22ec1d56e6"
      },
      "source": [
        "print(classification_report(y_test, y_cat_pred))"
      ],
      "execution_count": 54,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "              precision    recall  f1-score   support\n",
            "\n",
            "           0       0.86      0.88      0.87       923\n",
            "           1       0.85      0.89      0.87       864\n",
            "           2       0.98      0.57      0.72       148\n",
            "\n",
            "    accuracy                           0.86      1935\n",
            "   macro avg       0.89      0.78      0.82      1935\n",
            "weighted avg       0.86      0.86      0.86      1935\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "idXPeOixiPpI",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "b6d2cc31-306e-433e-eaf3-e73eb84eafd3"
      },
      "source": [
        "report_boot=bootstrap(np.array(X_test), np.array(y_test), tunned_model)\n",
        "\n",
        "for i in ['accuracy', 'macro avg', 'weighted avg']:\n",
        " if i == 'accuracy':\n",
        "     print(\"\\n\\n{:27}: {:.2f} ± {:.2f}\".format(i, report_boot[i][0], report_boot[i][1]))\n",
        " for j in ['f1-score', 'recall', 'precision']:\n",
        "   if i != 'accuracy':\n",
        "     print(\"{:15}  {:10}: {:.2f} ± {:.2f}\".format(i, j, report_boot[i][j][0], report_boot[i][j][1]))"
      ],
      "execution_count": 55,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "100%|██████████| 250/250 [02:28<00:00,  1.68it/s]"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "\n",
            "accuracy                   : 0.86 ± 0.01\n",
            "macro avg        f1-score  : 0.82 ± 0.01\n",
            "macro avg        recall    : 0.78 ± 0.01\n",
            "macro avg        precision : 0.89 ± 0.01\n",
            "weighted avg     f1-score  : 0.86 ± 0.01\n",
            "weighted avg     recall    : 0.86 ± 0.01\n",
            "weighted avg     precision : 0.86 ± 0.01\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "K8DHsrA_LvBf"
      },
      "source": [
        "# Usando o modelo Bert-NeuralMind"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7s2b1natMJMk"
      },
      "source": [
        "Um outro modelo pré-treinado que podemos usar do BERT é o modelo da NeuralMind, um centro de pesquisa de inteligência focada em ténicas avançadas de inteligência artificial, que foi treinado para o uso da língua portuguesa em geral.\n",
        "\n",
        "No Github do modelo [Solving classification problems with CatBoost](https://github.com/neuralmind-ai/portuguese-bert) pode ser encontrado um tutorial com experimentos com o CatBoost."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ye9DdtvGOhKE",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "56adff39-a953-4dd5-d17a-c96923a1a5fd"
      },
      "source": [
        "from legalnlp.get_premodel import *\n",
        "get_premodel('neuralmindbase')"
      ],
      "execution_count": 56,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "True"
            ]
          },
          "metadata": {},
          "execution_count": 56
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "84TIHrfQMKnn",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "e6693ea8-5f17-4fc0-83ec-b24234d54975"
      },
      "source": [
        "# Carregando o modelo da NeuralMind\n",
        "%%time\n",
        "\n",
        "bert_model =  BertModel.from_pretrained('/content/bert-base-portuguese-cased_pytorch_checkpoint').to(device)\n",
        "bert_tokenizer = BertTokenizer.from_pretrained('/content/vocab.txt', do_lower_case=False)\n",
        "\n",
        "clear_output()\n"
      ],
      "execution_count": 57,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "CPU times: user 2.17 s, sys: 438 ms, total: 2.61 s\n",
            "Wall time: 2.61 s\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dtP5eUbKKW8q",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "9ebe94ef-b446-404a-e933-1b9ceb57ecd3"
      },
      "source": [
        "# Fazendo a padronização dos textos\n",
        "wrapper = textwrap.TextWrapper()\n",
        "data_text = list(data['text'])\n",
        "\n",
        "for text in range(len(data_text[:4])):\n",
        "  print(f'{wrapper.fill(data_text[text])}')\n",
        "  print()"
      ],
      "execution_count": 58,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Início da Execução Juntado 0008003-11.2018.8.26.0002 - Cumprimento de\n",
            "sentença\n",
            "\n",
            "Ofício Requisitório - Comunicação - Protocolo Eletrônico - DEPRE\n",
            "\n",
            "Arquivado Provisoriamente aguardando manifestação\n",
            "\n",
            "Petição Juntada Juntada a petição diversa - Tipo: Petição\n",
            "Intermediária em Procedimento Comum Cível - Número: 80031 - Protocolo:\n",
            "FGRU19000140154 - Complemento: autor, expedição de ofício.\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Zbq2pNgZYWLv"
      },
      "source": [
        "# Aplicando o bert_tokenizer em nosso dataset com um comprimento máximo de 512 tokens\n",
        "encoded_inputs = bert_tokenizer(data_text, padding=True, truncation=True, max_length=512, return_tensors=\"pt\")\n",
        "\n",
        "#Agora temos nossos encoded_input em um dicionário com 3 chaves\n",
        "encoded_inputs.keys()\n",
        "input_ids = encoded_inputs['input_ids'].to(device)"
      ],
      "execution_count": 59,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "eQhdx-HoXs24",
        "outputId": "09f4d2d8-ad9a-4777-a902-bf16889040cb"
      },
      "source": [
        "# Criando o nosso vetor de features \n",
        "features = []\n",
        "\n",
        "# Aplicando o modelo pré-treinado em cada frase e adicionando-o ao nosso vetor\n",
        "\n",
        "for i in tqdm(range(len(data_text))):\n",
        "\n",
        "    with torch.no_grad():\n",
        "    \n",
        "      last_hidden_states = bert_model(input_ids[i:(i+1)])[1].cpu().numpy().reshape(-1).tolist()\n",
        "\n",
        "    features.append(last_hidden_states)\n"
      ],
      "execution_count": 60,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "100%|██████████| 6449/6449 [09:30<00:00, 11.31it/s]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "ZZV8FfAhbY_V",
        "outputId": "eebea2bf-8091-4b6c-baad-789d21c63f35"
      },
      "source": [
        "df_features = pd.DataFrame(features)\n",
        "features_label = pd.concat([df_features, data['encoded']], axis = 1)\n",
        "features_label.shape"
      ],
      "execution_count": 61,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(6449, 769)"
            ]
          },
          "metadata": {},
          "execution_count": 61
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Jq6s-6JEZAbv"
      },
      "source": [
        "# Dividir em treino e teste\n",
        "X_train, X_test, y_train, y_test = train_test_split(features_label.drop(columns = ['encoded']), features_label['encoded'], random_state = random_seed,test_size = 0.3)"
      ],
      "execution_count": 62,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "KaX3S6bVLvBj"
      },
      "source": [
        "## O Modelo de Boosting: CatBoost"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CfAkpMsFLvBk"
      },
      "source": [
        "# criando conjuntos de validação\n",
        "X_train, X_val, y_train, y_val = train_test_split(X_train, y_train, test_size = 0.1, stratify = y_train, random_state = random_seed)"
      ],
      "execution_count": 63,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "XBA3AWGDLvBk",
        "outputId": "4e68b759-ad6c-42b5-e15d-e371eeed4485"
      },
      "source": [
        "tunned_model = CatBoostClassifier(\n",
        "    loss_function = 'MultiClass',\n",
        "    random_seed=random_seed,\n",
        ")\n",
        "\n",
        "tunned_model.fit(\n",
        "    X_train, y_train,\n",
        "    verbose=500,\n",
        "    eval_set=(X_val, y_val),\n",
        "    early_stopping_rounds = 100\n",
        ")"
      ],
      "execution_count": 64,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Learning rate set to 0.111046\n",
            "0:\tlearn: 1.0018612\ttest: 1.0035581\tbest: 1.0035581 (0)\ttotal: 805ms\tremaining: 13m 24s\n",
            "500:\tlearn: 0.2243658\ttest: 0.4030074\tbest: 0.3991960 (405)\ttotal: 4m 18s\tremaining: 4m 17s\n",
            "Stopped by overfitting detector  (100 iterations wait)\n",
            "\n",
            "bestTest = 0.3991960311\n",
            "bestIteration = 405\n",
            "\n",
            "Shrink model to first 406 iterations.\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<catboost.core.CatBoostClassifier at 0x7f87b917c610>"
            ]
          },
          "metadata": {},
          "execution_count": 64
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Vs62EcCOLvBk"
      },
      "source": [
        "y_cat_pred = tunned_model.predict(X_test)"
      ],
      "execution_count": 65,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "bTPnGEvfLvBk",
        "outputId": "98c88fec-d4dc-4609-8373-5250b91bdce4"
      },
      "source": [
        "accuracy_score(y_test, y_cat_pred)"
      ],
      "execution_count": 66,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0.8382428940568476"
            ]
          },
          "metadata": {},
          "execution_count": 66
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "pW67UsZ8LvBk",
        "outputId": "51a6eaaa-e639-4fcd-fe98-8ab012ad1b2c"
      },
      "source": [
        "tunned_model.predict_proba(X_test)[:5]"
      ],
      "execution_count": 67,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([[0.57062936, 0.37211143, 0.05725921],\n",
              "       [0.2342113 , 0.68459205, 0.08119665],\n",
              "       [0.0556465 , 0.93924415, 0.00510935],\n",
              "       [0.9708583 , 0.0248357 , 0.00430599],\n",
              "       [0.14835578, 0.7349196 , 0.11672462]])"
            ]
          },
          "metadata": {},
          "execution_count": 67
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 301
        },
        "id": "SDGxxyyZLvBk",
        "outputId": "170f32a4-21e7-42cd-8720-a3a359fd9606"
      },
      "source": [
        "cm = confusion_matrix(y_test, y_cat_pred)\n",
        "sns.heatmap(cm, annot = True, cmap = 'Greens', fmt = '.3g')\n",
        "plt.title(\"Heatmap\")\n",
        "plt.show()"
      ],
      "execution_count": 68,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "l7_lodZeLvBl",
        "outputId": "24d481fc-dd5d-4a64-834d-7a427b989a36"
      },
      "source": [
        "print(classification_report(y_test, y_cat_pred))"
      ],
      "execution_count": 69,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "              precision    recall  f1-score   support\n",
            "\n",
            "           0       0.84      0.86      0.85       923\n",
            "           1       0.83      0.87      0.85       864\n",
            "           2       0.96      0.49      0.65       148\n",
            "\n",
            "    accuracy                           0.84      1935\n",
            "   macro avg       0.88      0.74      0.78      1935\n",
            "weighted avg       0.84      0.84      0.83      1935\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "oDMSIhazm0JP",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "71717f94-3612-4f89-9e80-2f42a67c8cf5"
      },
      "source": [
        "report_boot=bootstrap(np.array(X_test), np.array(y_test), tunned_model)\n",
        "\n",
        "for i in ['accuracy', 'macro avg', 'weighted avg']:\n",
        " if i == 'accuracy':\n",
        "     print(\"\\n\\n{:27}: {:.2f} ± {:.2f}\".format(i, report_boot[i][0], report_boot[i][1]))\n",
        " for j in ['f1-score', 'recall', 'precision']:\n",
        "   if i != 'accuracy':\n",
        "     print(\"{:15}  {:10}: {:.2f} ± {:.2f}\".format(i, j, report_boot[i][j][0], report_boot[i][j][1]))"
      ],
      "execution_count": 70,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "100%|██████████| 250/250 [02:28<00:00,  1.68it/s]"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "\n",
            "accuracy                   : 0.84 ± 0.01\n",
            "macro avg        f1-score  : 0.78 ± 0.01\n",
            "macro avg        recall    : 0.74 ± 0.01\n",
            "macro avg        precision : 0.88 ± 0.01\n",
            "weighted avg     f1-score  : 0.83 ± 0.01\n",
            "weighted avg     recall    : 0.84 ± 0.01\n",
            "weighted avg     precision : 0.84 ± 0.01\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "YwNE19klSXsY"
      },
      "source": [
        "#Referências"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "YUkcO9NySaWw"
      },
      "source": [
        "[1] Devlin J., Chang M., Lee K., Toutanova K. (2019). BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. arxiv: 1810.04805. \\\\\n",
        "\n",
        "[2] Vaswani A., Shazeer N., Parmar N., et al. (2017). Attention Is All You Need. arxiv: 1706.03762. \\\\\n",
        "\n",
        "[3] Devlin J. Google AI Language. BERT: Pre-training of Deep Bidirectional\n",
        "Transformers for Language Understanding (Bidirectional Encoder Representations from Transformers). Stanford. https://nlp.stanford.edu/seminar/details/jdevlin.pdf. Acesso em: 04/08/2021. \\\\\n",
        "\n",
        "[4] Alammar J., The Illustrated BERT, ELMo, and co. (How NLP Cracked Transfer Learning). https://jalammar.github.io/illustrated-bert/. Acesso em: 04/08/2021. \\\\\n",
        "\n",
        "[5] Horev R. (2018). BERT Explained: State of the art language model for NLP. https://towardsdatascience.com/bert-explained-state-of-the-art-language-model-for-nlp-f8b21a9b6270. Acesso em: 04/08/2021. \\\\\n",
        "\n",
        "[6] McCormick C., Ryan N. BERT Fine-Tuning Tutorial with PyTorch. https://colab.research.google.com/drive/1pTuQhug6Dhl9XalKB0zUGf4FIdYFlpcX#scrollTo=9cQNvaZ9bnyy. Acesso em: 04/08/2021. \\\\\n",
        "\n",
        "[7] McCormick C. Question Answering with a Fine-Tuned BERT. https://mccormickml.com/2020/03/10/question-answering-with-a-fine-tuned-BERT/. Acesso em: 04/08/2021.  "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "sUy1f81TpRnm"
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    }
  ]
}