5 семестр/МО/ЛР-5.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7YUvcm5Vte58"
      },
      "source": [
        "# Лабораторная работа №5\n",
        "> Выполнили:  \n",
        "> – 339080, Дьячихин Данила Никитич  \n",
        "> – 368606, Овчинников Павел Алексеевич\n",
        "\n",
        "**Цель работы:** построить модели классификации на основе ансамблевых алгоритмов: бэггинг и градиентный бустинг, содержащие в качестве базовых алгоритмов решающие деревья.\n",
        "\n",
        "**Инструменты:** бэггинг решающих деревьев (случайный лес, `RandomForest`), градиентный бустинг над решающими деревьями (`GradientBoostingRegressor`).\n",
        "\n",
        "Будем использовать датасет, полученный в результате выполнения лабораторной работы №1 — `heart_preprocessed.csv`. Использование `Advertising_preprocessed.csv` опустим, т.к. этот датасет не является классическим для задач классификации, в отличие от `heart_preprocessed.csv`."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "QunW8Oi2te6A"
      },
      "source": [
        "Импортируем необходимые библиотеки и загрузим датасеты"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 2,
      "metadata": {
        "id": "UqK0SERXte6B",
        "outputId": "c3358387-5db7-41df-c1f5-291049c75c9b",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "<class 'pandas.core.frame.DataFrame'>\n",
            "RangeIndex: 296 entries, 0 to 295\n",
            "Data columns (total 14 columns):\n",
            " #   Column    Non-Null Count  Dtype  \n",
            "---  ------    --------------  -----  \n",
            " 0   age       296 non-null    float64\n",
            " 1   sex       296 non-null    int64  \n",
            " 2   cp        296 non-null    int64  \n",
            " 3   trestbps  296 non-null    float64\n",
            " 4   chol      296 non-null    float64\n",
            " 5   fbs       296 non-null    int64  \n",
            " 6   restecg   296 non-null    int64  \n",
            " 7   thalach   296 non-null    float64\n",
            " 8   exang     296 non-null    int64  \n",
            " 9   oldpeak   296 non-null    float64\n",
            " 10  slope     296 non-null    int64  \n",
            " 11  ca        296 non-null    int64  \n",
            " 12  thal      296 non-null    int64  \n",
            " 13  target    296 non-null    int64  \n",
            "dtypes: float64(5), int64(9)\n",
            "memory usage: 32.5 KB\n",
            "None\n"
          ]
        }
      ],
      "source": [
        "import matplotlib.pyplot as plt\n",
        "import pandas as pd\n",
        "from seaborn import heatmap\n",
        "from sklearn.ensemble import GradientBoostingClassifier, RandomForestClassifier\n",
        "from sklearn.metrics import classification_report, confusion_matrix, f1_score\n",
        "from sklearn.model_selection import GridSearchCV, train_test_split\n",
        "\n",
        "heart = pd.read_csv('heart_preprocessed.csv')\n",
        "print(heart.info())"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mH_8q7mZte6E"
      },
      "source": [
        "Теперь разделим датасеты на признаки и целевую переменную, и обучающую и тестовую выборки в соотношении 80/20."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 3,
      "metadata": {
        "id": "hrqzXyz8te6E"
      },
      "outputs": [],
      "source": [
        "X = heart.drop('target', axis=1)\n",
        "Y = heart['target']\n",
        "X_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=0.2, random_state=42)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "9SHSWBPPte6F"
      },
      "source": [
        "# Бэггинг решающих деревьев (случайный лес, `RandomForest`)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vrGENGhote6G"
      },
      "source": [
        "Зададим параметры для перебора в `GridSearchCV` и обучим модель случайного леса на лучших параметрах."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 4,
      "metadata": {
        "id": "V0VZYUqxte6H"
      },
      "outputs": [],
      "source": [
        "classifier = RandomForestClassifier(random_state=42)\n",
        "params = {\n",
        "    'n_estimators': [100, 200, 300, 400],\n",
        "    'max_depth': [None, 5, 10, 20],\n",
        "    'min_samples_split': [2, 5, 10, 20],\n",
        "    'min_samples_leaf': [1, 2, 4, 8],\n",
        "}\n",
        "grid_search = GridSearchCV(estimator=classifier, param_grid=params, scoring='accuracy')\n",
        "\n",
        "grid_search.fit(X_train, Y_train)\n",
        "y_predict_rfc = grid_search.best_estimator_.predict(X_test)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Xwv_Mj-Rte6H"
      },
      "source": [
        "Теперь выведем отчёт о классификации на тестовой выборке, чтобы оценить её качество."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 5,
      "metadata": {
        "id": "WYVUID6Gte6I",
        "outputId": "471175a9-0d20-48d9-92cb-0a62ddaaf256",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Отчет о классификации:\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "           0       0.83      0.86      0.84        28\n",
            "           1       0.87      0.84      0.86        32\n",
            "\n",
            "    accuracy                           0.85        60\n",
            "   macro avg       0.85      0.85      0.85        60\n",
            "weighted avg       0.85      0.85      0.85        60\n",
            "\n",
            "F1 (f1_score):  0.8571428571428571\n"
          ]
        }
      ],
      "source": [
        "print('Отчет о классификации:', classification_report(y_true=Y_test, y_pred=y_predict_rfc), sep='\\n')\n",
        "print('F1 (f1_score): ', f1_score(y_true=Y_test, y_pred=y_predict_rfc))"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "RqMxYjHVte6I"
      },
      "source": [
        "# Градиентный бустинг над решающими деревьями (`GradientBoostingRegressor`)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ZGis1rD4te6I"
      },
      "source": [
        "Точно так же зададим параметры для перебора в `GridSearchCV` и обучим модель градиентного бустинга на лучших параметрах, которые выберет `GridSearchCV`."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 6,
      "metadata": {
        "id": "pD0e59ujte6J"
      },
      "outputs": [],
      "source": [
        "classifier = GradientBoostingClassifier(random_state=42)\n",
        "params = {\n",
        "    'n_estimators': [50, 100, 200, 400],\n",
        "    'learning_rate': [0.01, 0.1, 0.2, 0.5],\n",
        "    'max_depth': [3, 4, 5, 6],\n",
        "}\n",
        "\n",
        "grid_search = GridSearchCV(estimator=classifier, param_grid=params, scoring='accuracy')\n",
        "\n",
        "grid_search.fit(X_train, Y_train)\n",
        "y_predict_gbr = grid_search.best_estimator_.predict(X_test)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "otKTkrUFte6J"
      },
      "source": [
        "Выведем для модели отчёт о классификации на тестовой выборке, чтобы оценить её качество."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 7,
      "metadata": {
        "id": "LP1TeY7Bte6K",
        "outputId": "f88fb7a0-d95e-4a0c-b1be-7b6cb747e9a9",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Отчет о классификации:\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "           0       0.85      0.82      0.84        28\n",
            "           1       0.85      0.88      0.86        32\n",
            "\n",
            "    accuracy                           0.85        60\n",
            "   macro avg       0.85      0.85      0.85        60\n",
            "weighted avg       0.85      0.85      0.85        60\n",
            "\n",
            "F1 (f1_score):  0.8615384615384616\n"
          ]
        }
      ],
      "source": [
        "print('Отчет о классификации:', classification_report(y_true=Y_test, y_pred=y_predict_gbr), sep='\\n')\n",
        "print('F1 (f1_score): ', f1_score(y_true=Y_test, y_pred=y_predict_gbr))"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "J9UrieFste6L"
      },
      "source": [
        "# Сравнительная оценка качества моделей"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "x4k4Epovte6L"
      },
      "source": [
        "Определим критерии сравнения моделей:\n",
        "- `f1_score` — среднее гармоническое между точностью и полнотой\n",
        "- `heatmap` — матрица ошибок\n",
        "\n",
        "Сравним модели по этим критериям и сделаем вывод о качестве работы моделей."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 8,
      "metadata": {
        "id": "5T5MzHfOte6L",
        "outputId": "61cf8f8e-e8c5-4317-c6d1-aadcdff436db",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 442
        }
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "f1_score RandomForestClassifier:  0.8571428571428571\n",
            "f1_score GradientBoostingClassifier:  0.8615384615384616\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 900x400 with 4 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ],
      "source": [
        "print('f1_score RandomForestClassifier: ', f1_score(y_true=Y_test, y_pred=y_predict_rfc))\n",
        "print('f1_score GradientBoostingClassifier: ', f1_score(y_true=Y_test, y_pred=y_predict_gbr))\n",
        "\n",
        "plt.figure(figsize=(9, 4))\n",
        "plt.subplot(1, 2, 1)\n",
        "plt.title('RandomForestClassifier')\n",
        "heatmap(confusion_matrix(Y_test, y_predict_rfc), annot=True, fmt='d', cmap='Blues')\n",
        "plt.subplot(1, 2, 2)\n",
        "plt.title('GradientBoostingClassifier')\n",
        "heatmap(confusion_matrix(Y_test, y_predict_gbr), annot=True, fmt='d', cmap='Blues')\n",
        "plt.tight_layout()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "gbsbhmknte6M"
      },
      "source": [
        "# Вывод\n",
        "Как мы видим, `f1_score` у обеих моделей примерно одинаковый. Матрица ошибок показывает, что `RandomForest` ошибся на классе 1 на один раз больше, а `GradientBoostingRegressor` — точно так же на классе 0. В целом, обе модели показывают себя хорошо, но `GradientBoostingRegressor` работает гораздо быстрее, чем `RandomForest` (118 против 553 секунд), поэтому предпочтительнее использовать именно его, т.к. цена ошибки не такая большая относительно времени работы."
      ]
    }
  ],
  "metadata": {
    "kernelspec": {
      "display_name": "education",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.12.7"
    },
    "colab": {
      "provenance": []
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}