1.1/notebooks/auto_examples/preprocessing/plot_discretization_classification.ipynb

{
  "cells": [
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "collapsed": false
      },
      "outputs": [],
      "source": [
        "%matplotlib inline"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "\n# Feature discretization\n\nA demonstration of feature discretization on synthetic classification datasets.\nFeature discretization decomposes each feature into a set of bins, here equally\ndistributed in width. The discrete values are then one-hot encoded, and given\nto a linear classifier. This preprocessing enables a non-linear behavior even\nthough the classifier is linear.\n\nOn this example, the first two rows represent linearly non-separable datasets\n(moons and concentric circles) while the third is approximately linearly\nseparable. On the two linearly non-separable datasets, feature discretization\nlargely increases the performance of linear classifiers. On the linearly\nseparable dataset, feature discretization decreases the performance of linear\nclassifiers. Two non-linear classifiers are also shown for comparison.\n\nThis example should be taken with a grain of salt, as the intuition conveyed\ndoes not necessarily carry over to real datasets. Particularly in\nhigh-dimensional spaces, data can more easily be separated linearly. Moreover,\nusing feature discretization and one-hot encoding increases the number of\nfeatures, which easily lead to overfitting when the number of samples is small.\n\nThe plots show training points in solid colors and testing points\nsemi-transparent. The lower right shows the classification accuracy on the test\nset.\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "collapsed": false
      },
      "outputs": [],
      "source": [
        "# Code source: Tom Dupr\u00e9 la Tour\n# Adapted from plot_classifier_comparison by Ga\u00ebl Varoquaux and Andreas M\u00fcller\n#\n# License: BSD 3 clause\n\nimport numpy as np\nimport matplotlib.pyplot as plt\nfrom matplotlib.colors import ListedColormap\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.preprocessing import StandardScaler\nfrom sklearn.datasets import make_moons, make_circles, make_classification\nfrom sklearn.linear_model import LogisticRegression\nfrom sklearn.model_selection import GridSearchCV\nfrom sklearn.pipeline import make_pipeline\nfrom sklearn.preprocessing import KBinsDiscretizer\nfrom sklearn.svm import SVC, LinearSVC\nfrom sklearn.ensemble import GradientBoostingClassifier\nfrom sklearn.utils._testing import ignore_warnings\nfrom sklearn.exceptions import ConvergenceWarning\n\nh = 0.02  # step size in the mesh\n\n\ndef get_name(estimator):\n    name = estimator.__class__.__name__\n    if name == \"Pipeline\":\n        name = [get_name(est[1]) for est in estimator.steps]\n        name = \" + \".join(name)\n    return name\n\n\n# list of (estimator, param_grid), where param_grid is used in GridSearchCV\n# The parameter spaces in this example are limited to a narrow band to reduce\n# its runtime. In a real use case, a broader search space for the algorithms\n# should be used.\nclassifiers = [\n    (\n        make_pipeline(StandardScaler(), LogisticRegression(random_state=0)),\n        {\"logisticregression__C\": np.logspace(-1, 1, 3)},\n    ),\n    (\n        make_pipeline(StandardScaler(), LinearSVC(random_state=0)),\n        {\"linearsvc__C\": np.logspace(-1, 1, 3)},\n    ),\n    (\n        make_pipeline(\n            StandardScaler(),\n            KBinsDiscretizer(encode=\"onehot\"),\n            LogisticRegression(random_state=0),\n        ),\n        {\n            \"kbinsdiscretizer__n_bins\": np.arange(5, 8),\n            \"logisticregression__C\": np.logspace(-1, 1, 3),\n        },\n    ),\n    (\n        make_pipeline(\n            StandardScaler(),\n            KBinsDiscretizer(encode=\"onehot\"),\n            LinearSVC(random_state=0),\n        ),\n        {\n            \"kbinsdiscretizer__n_bins\": np.arange(5, 8),\n            \"linearsvc__C\": np.logspace(-1, 1, 3),\n        },\n    ),\n    (\n        make_pipeline(\n            StandardScaler(), GradientBoostingClassifier(n_estimators=5, random_state=0)\n        ),\n        {\"gradientboostingclassifier__learning_rate\": np.logspace(-2, 0, 5)},\n    ),\n    (\n        make_pipeline(StandardScaler(), SVC(random_state=0)),\n        {\"svc__C\": np.logspace(-1, 1, 3)},\n    ),\n]\n\nnames = [get_name(e).replace(\"StandardScaler + \", \"\") for e, _ in classifiers]\n\nn_samples = 100\ndatasets = [\n    make_moons(n_samples=n_samples, noise=0.2, random_state=0),\n    make_circles(n_samples=n_samples, noise=0.2, factor=0.5, random_state=1),\n    make_classification(\n        n_samples=n_samples,\n        n_features=2,\n        n_redundant=0,\n        n_informative=2,\n        random_state=2,\n        n_clusters_per_class=1,\n    ),\n]\n\nfig, axes = plt.subplots(\n    nrows=len(datasets), ncols=len(classifiers) + 1, figsize=(21, 9)\n)\n\ncm_piyg = plt.cm.PiYG\ncm_bright = ListedColormap([\"#b30065\", \"#178000\"])\n\n# iterate over datasets\nfor ds_cnt, (X, y) in enumerate(datasets):\n    print(f\"\\ndataset {ds_cnt}\\n---------\")\n\n    # split into training and test part\n    X_train, X_test, y_train, y_test = train_test_split(\n        X, y, test_size=0.5, random_state=42\n    )\n\n    # create the grid for background colors\n    x_min, x_max = X[:, 0].min() - 0.5, X[:, 0].max() + 0.5\n    y_min, y_max = X[:, 1].min() - 0.5, X[:, 1].max() + 0.5\n    xx, yy = np.meshgrid(np.arange(x_min, x_max, h), np.arange(y_min, y_max, h))\n\n    # plot the dataset first\n    ax = axes[ds_cnt, 0]\n    if ds_cnt == 0:\n        ax.set_title(\"Input data\")\n    # plot the training points\n    ax.scatter(X_train[:, 0], X_train[:, 1], c=y_train, cmap=cm_bright, edgecolors=\"k\")\n    # and testing points\n    ax.scatter(\n        X_test[:, 0], X_test[:, 1], c=y_test, cmap=cm_bright, alpha=0.6, edgecolors=\"k\"\n    )\n    ax.set_xlim(xx.min(), xx.max())\n    ax.set_ylim(yy.min(), yy.max())\n    ax.set_xticks(())\n    ax.set_yticks(())\n\n    # iterate over classifiers\n    for est_idx, (name, (estimator, param_grid)) in enumerate(zip(names, classifiers)):\n        ax = axes[ds_cnt, est_idx + 1]\n\n        clf = GridSearchCV(estimator=estimator, param_grid=param_grid)\n        with ignore_warnings(category=ConvergenceWarning):\n            clf.fit(X_train, y_train)\n        score = clf.score(X_test, y_test)\n        print(f\"{name}: {score:.2f}\")\n\n        # plot the decision boundary. For that, we will assign a color to each\n        # point in the mesh [x_min, x_max]*[y_min, y_max].\n        if hasattr(clf, \"decision_function\"):\n            Z = clf.decision_function(np.column_stack([xx.ravel(), yy.ravel()]))\n        else:\n            Z = clf.predict_proba(np.column_stack([xx.ravel(), yy.ravel()]))[:, 1]\n\n        # put the result into a color plot\n        Z = Z.reshape(xx.shape)\n        ax.contourf(xx, yy, Z, cmap=cm_piyg, alpha=0.8)\n\n        # plot the training points\n        ax.scatter(\n            X_train[:, 0], X_train[:, 1], c=y_train, cmap=cm_bright, edgecolors=\"k\"\n        )\n        # and testing points\n        ax.scatter(\n            X_test[:, 0],\n            X_test[:, 1],\n            c=y_test,\n            cmap=cm_bright,\n            edgecolors=\"k\",\n            alpha=0.6,\n        )\n        ax.set_xlim(xx.min(), xx.max())\n        ax.set_ylim(yy.min(), yy.max())\n        ax.set_xticks(())\n        ax.set_yticks(())\n\n        if ds_cnt == 0:\n            ax.set_title(name.replace(\" + \", \"\\n\"))\n        ax.text(\n            0.95,\n            0.06,\n            (f\"{score:.2f}\").lstrip(\"0\"),\n            size=15,\n            bbox=dict(boxstyle=\"round\", alpha=0.8, facecolor=\"white\"),\n            transform=ax.transAxes,\n            horizontalalignment=\"right\",\n        )\n\n\nplt.tight_layout()\n\n# Add suptitles above the figure\nplt.subplots_adjust(top=0.90)\nsuptitles = [\n    \"Linear classifiers\",\n    \"Feature discretization and linear classifiers\",\n    \"Non-linear classifiers\",\n]\nfor i, suptitle in zip([1, 3, 5], suptitles):\n    ax = axes[0, i]\n    ax.text(\n        1.05,\n        1.25,\n        suptitle,\n        transform=ax.transAxes,\n        horizontalalignment=\"center\",\n        size=\"x-large\",\n    )\nplt.show()"
      ]
    }
  ],
  "metadata": {
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.9.13"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}