{ "cells": [ { "cell_type": "markdown", "metadata": {}, "source": [ "**Chapter 5 – Support Vector Machines**\n", "\n", "_This notebook contains all the sample code and solutions to the exercises in chapter 5._" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "\n", " \n", "
\n", " Run in Google Colab\n", "
" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# Setup" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "First, let's import a few common modules, ensure MatplotLib plots figures inline and prepare a function to save the figures. We also check that Python 3.5 or later is installed (although Python 2.x may work, it is deprecated so we strongly recommend you use Python 3 instead), as well as Scikit-Learn ≥0.20." ] }, { "cell_type": "code", "execution_count": 1, "metadata": {}, "outputs": [], "source": [ "# Python ≥3.5 is required\n", "import sys\n", "assert sys.version_info >= (3, 5)\n", "\n", "# Scikit-Learn ≥0.20 is required\n", "import sklearn\n", "assert sklearn.__version__ >= \"0.20\"\n", "\n", "# Common imports\n", "import numpy as np\n", "import os\n", "\n", "# to make this notebook's output stable across runs\n", "np.random.seed(42)\n", "\n", "# To plot pretty figures\n", "%matplotlib inline\n", "import matplotlib as mpl\n", "import matplotlib.pyplot as plt\n", "mpl.rc('axes', labelsize=14)\n", "mpl.rc('xtick', labelsize=12)\n", "mpl.rc('ytick', labelsize=12)\n", "\n", "# Where to save the figures\n", "PROJECT_ROOT_DIR = \".\"\n", "CHAPTER_ID = \"svm\"\n", "IMAGES_PATH = os.path.join(PROJECT_ROOT_DIR, \"images\", CHAPTER_ID)\n", "os.makedirs(IMAGES_PATH, exist_ok=True)\n", "\n", "def save_fig(fig_id, tight_layout=True, fig_extension=\"png\", resolution=300):\n", " path = os.path.join(IMAGES_PATH, fig_id + \".\" + fig_extension)\n", " print(\"Saving figure\", fig_id)\n", " if tight_layout:\n", " plt.tight_layout()\n", " plt.savefig(path, format=fig_extension, dpi=resolution)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# Large margin classification" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "The next few code cells generate the first figures in chapter 5. The first actual code sample comes after:" ] }, { "cell_type": "code", "execution_count": 2, "metadata": {}, "outputs": [], "source": [ "from sklearn.svm import SVC\n", "from sklearn import datasets\n", "\n", "iris = datasets.load_iris()\n", "X = iris[\"data\"][:, (2, 3)] # petal length, petal width\n", "y = iris[\"target\"]\n", "\n", "setosa_or_versicolor = (y == 0) | (y == 1)\n", "X = X[setosa_or_versicolor]\n", "y = y[setosa_or_versicolor]\n", "\n", "# SVM Classifier model\n", "svm_clf = SVC(kernel=\"linear\", C=float(\"inf\"))\n", "svm_clf.fit(X, y)" ] }, { "cell_type": "code", "execution_count": 3, "metadata": {}, "outputs": [], "source": [ "# Bad models\n", "x0 = np.linspace(0, 5.5, 200)\n", "pred_1 = 5*x0 - 20\n", "pred_2 = x0 - 1.8\n", "pred_3 = 0.1 * x0 + 0.5\n", "\n", "def plot_svc_decision_boundary(svm_clf, xmin, xmax):\n", " w = svm_clf.coef_[0]\n", " b = svm_clf.intercept_[0]\n", "\n", " # At the decision boundary, w0*x0 + w1*x1 + b = 0\n", " # => x1 = -w0/w1 * x0 - b/w1\n", " x0 = np.linspace(xmin, xmax, 200)\n", " decision_boundary = -w[0]/w[1] * x0 - b/w[1]\n", "\n", " margin = 1/w[1]\n", " gutter_up = decision_boundary + margin\n", " gutter_down = decision_boundary - margin\n", "\n", " svs = svm_clf.support_vectors_\n", " plt.scatter(svs[:, 0], svs[:, 1], s=180, facecolors='#FFAAAA')\n", " plt.plot(x0, decision_boundary, \"k-\", linewidth=2)\n", " plt.plot(x0, gutter_up, \"k--\", linewidth=2)\n", " plt.plot(x0, gutter_down, \"k--\", linewidth=2)\n", "\n", "fig, axes = plt.subplots(ncols=2, figsize=(10,2.7), sharey=True)\n", "\n", "plt.sca(axes[0])\n", "plt.plot(x0, pred_1, \"g--\", linewidth=2)\n", "plt.plot(x0, pred_2, \"m-\", linewidth=2)\n", "plt.plot(x0, pred_3, \"r-\", linewidth=2)\n", "plt.plot(X[:, 0][y==1], X[:, 1][y==1], \"bs\", label=\"Iris versicolor\")\n", "plt.plot(X[:, 0][y==0], X[:, 1][y==0], \"yo\", label=\"Iris setosa\")\n", "plt.xlabel(\"Petal length\", fontsize=14)\n", "plt.ylabel(\"Petal width\", fontsize=14)\n", "plt.legend(loc=\"upper left\", fontsize=14)\n", "plt.axis([0, 5.5, 0, 2])\n", "\n", "plt.sca(axes[1])\n", "plot_svc_decision_boundary(svm_clf, 0, 5.5)\n", "plt.plot(X[:, 0][y==1], X[:, 1][y==1], \"bs\")\n", "plt.plot(X[:, 0][y==0], X[:, 1][y==0], \"yo\")\n", "plt.xlabel(\"Petal length\", fontsize=14)\n", "plt.axis([0, 5.5, 0, 2])\n", "\n", "save_fig(\"large_margin_classification_plot\")\n", "plt.show()" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# Sensitivity to feature scales" ] }, { "cell_type": "code", "execution_count": 4, "metadata": {}, "outputs": [], "source": [ "Xs = np.array([[1, 50], [5, 20], [3, 80], [5, 60]]).astype(np.float64)\n", "ys = np.array([0, 0, 1, 1])\n", "svm_clf = SVC(kernel=\"linear\", C=100)\n", "svm_clf.fit(Xs, ys)\n", "\n", "plt.figure(figsize=(9,2.7))\n", "plt.subplot(121)\n", "plt.plot(Xs[:, 0][ys==1], Xs[:, 1][ys==1], \"bo\")\n", "plt.plot(Xs[:, 0][ys==0], Xs[:, 1][ys==0], \"ms\")\n", "plot_svc_decision_boundary(svm_clf, 0, 6)\n", "plt.xlabel(\"$x_0$\", fontsize=20)\n", "plt.ylabel(\"$x_1$    \", fontsize=20, rotation=0)\n", "plt.title(\"Unscaled\", fontsize=16)\n", "plt.axis([0, 6, 0, 90])\n", "\n", "from sklearn.preprocessing import StandardScaler\n", "scaler = StandardScaler()\n", "X_scaled = scaler.fit_transform(Xs)\n", "svm_clf.fit(X_scaled, ys)\n", "\n", "plt.subplot(122)\n", "plt.plot(X_scaled[:, 0][ys==1], X_scaled[:, 1][ys==1], \"bo\")\n", "plt.plot(X_scaled[:, 0][ys==0], X_scaled[:, 1][ys==0], \"ms\")\n", "plot_svc_decision_boundary(svm_clf, -2, 2)\n", "plt.xlabel(\"$x_0$\", fontsize=20)\n", "plt.ylabel(\"$x'_1$ \", fontsize=20, rotation=0)\n", "plt.title(\"Scaled\", fontsize=16)\n", "plt.axis([-2, 2, -2, 2])\n", "\n", "save_fig(\"sensitivity_to_feature_scales_plot\")\n" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# Sensitivity to outliers" ] }, { "cell_type": "code", "execution_count": 5, "metadata": {}, "outputs": [], "source": [ "X_outliers = np.array([[3.4, 1.3], [3.2, 0.8]])\n", "y_outliers = np.array([0, 0])\n", "Xo1 = np.concatenate([X, X_outliers[:1]], axis=0)\n", "yo1 = np.concatenate([y, y_outliers[:1]], axis=0)\n", "Xo2 = np.concatenate([X, X_outliers[1:]], axis=0)\n", "yo2 = np.concatenate([y, y_outliers[1:]], axis=0)\n", "\n", "svm_clf2 = SVC(kernel=\"linear\", C=10**9)\n", "svm_clf2.fit(Xo2, yo2)\n", "\n", "fig, axes = plt.subplots(ncols=2, figsize=(10,2.7), sharey=True)\n", "\n", "plt.sca(axes[0])\n", "plt.plot(Xo1[:, 0][yo1==1], Xo1[:, 1][yo1==1], \"bs\")\n", "plt.plot(Xo1[:, 0][yo1==0], Xo1[:, 1][yo1==0], \"yo\")\n", "plt.text(0.3, 1.0, \"Impossible!\", fontsize=24, color=\"red\")\n", "plt.xlabel(\"Petal length\", fontsize=14)\n", "plt.ylabel(\"Petal width\", fontsize=14)\n", "plt.annotate(\"Outlier\",\n", " xy=(X_outliers[0][0], X_outliers[0][1]),\n", " xytext=(2.5, 1.7),\n", " ha=\"center\",\n", " arrowprops=dict(facecolor='black', shrink=0.1),\n", " fontsize=16,\n", " )\n", "plt.axis([0, 5.5, 0, 2])\n", "\n", "plt.sca(axes[1])\n", "plt.plot(Xo2[:, 0][yo2==1], Xo2[:, 1][yo2==1], \"bs\")\n", "plt.plot(Xo2[:, 0][yo2==0], Xo2[:, 1][yo2==0], \"yo\")\n", "plot_svc_decision_boundary(svm_clf2, 0, 5.5)\n", "plt.xlabel(\"Petal length\", fontsize=14)\n", "plt.annotate(\"Outlier\",\n", " xy=(X_outliers[1][0], X_outliers[1][1]),\n", " xytext=(3.2, 0.08),\n", " ha=\"center\",\n", " arrowprops=dict(facecolor='black', shrink=0.1),\n", " fontsize=16,\n", " )\n", "plt.axis([0, 5.5, 0, 2])\n", "\n", "save_fig(\"sensitivity_to_outliers_plot\")\n", "plt.show()" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# Large margin *vs* margin violations" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "This is the first code example in chapter 5:" ] }, { "cell_type": "code", "execution_count": 6, "metadata": {}, "outputs": [], "source": [ "import numpy as np\n", "from sklearn import datasets\n", "from sklearn.pipeline import Pipeline\n", "from sklearn.preprocessing import StandardScaler\n", "from sklearn.svm import LinearSVC\n", "\n", "iris = datasets.load_iris()\n", "X = iris[\"data\"][:, (2, 3)] # petal length, petal width\n", "y = (iris[\"target\"] == 2).astype(np.float64) # Iris virginica\n", "\n", "svm_clf = Pipeline([\n", " (\"scaler\", StandardScaler()),\n", " (\"linear_svc\", LinearSVC(C=1, loss=\"hinge\", random_state=42)),\n", " ])\n", "\n", "svm_clf.fit(X, y)" ] }, { "cell_type": "code", "execution_count": 7, "metadata": {}, "outputs": [], "source": [ "svm_clf.predict([[5.5, 1.7]])" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Now let's generate the graph comparing different regularization settings:" ] }, { "cell_type": "code", "execution_count": 8, "metadata": {}, "outputs": [], "source": [ "scaler = StandardScaler()\n", "svm_clf1 = LinearSVC(C=1, loss=\"hinge\", random_state=42)\n", "svm_clf2 = LinearSVC(C=100, loss=\"hinge\", random_state=42)\n", "\n", "scaled_svm_clf1 = Pipeline([\n", " (\"scaler\", scaler),\n", " (\"linear_svc\", svm_clf1),\n", " ])\n", "scaled_svm_clf2 = Pipeline([\n", " (\"scaler\", scaler),\n", " (\"linear_svc\", svm_clf2),\n", " ])\n", "\n", "scaled_svm_clf1.fit(X, y)\n", "scaled_svm_clf2.fit(X, y)" ] }, { "cell_type": "code", "execution_count": 9, "metadata": {}, "outputs": [], "source": [ "# Convert to unscaled parameters\n", "b1 = svm_clf1.decision_function([-scaler.mean_ / scaler.scale_])\n", "b2 = svm_clf2.decision_function([-scaler.mean_ / scaler.scale_])\n", "w1 = svm_clf1.coef_[0] / scaler.scale_\n", "w2 = svm_clf2.coef_[0] / scaler.scale_\n", "svm_clf1.intercept_ = np.array([b1])\n", "svm_clf2.intercept_ = np.array([b2])\n", "svm_clf1.coef_ = np.array([w1])\n", "svm_clf2.coef_ = np.array([w2])\n", "\n", "# Find support vectors (LinearSVC does not do this automatically)\n", "t = y * 2 - 1\n", "support_vectors_idx1 = (t * (X.dot(w1) + b1) < 1).ravel()\n", "support_vectors_idx2 = (t * (X.dot(w2) + b2) < 1).ravel()\n", "svm_clf1.support_vectors_ = X[support_vectors_idx1]\n", "svm_clf2.support_vectors_ = X[support_vectors_idx2]" ] }, { "cell_type": "code", "execution_count": 10, "metadata": {}, "outputs": [], "source": [ "fig, axes = plt.subplots(ncols=2, figsize=(10,2.7), sharey=True)\n", "\n", "plt.sca(axes[0])\n", "plt.plot(X[:, 0][y==1], X[:, 1][y==1], \"g^\", label=\"Iris virginica\")\n", "plt.plot(X[:, 0][y==0], X[:, 1][y==0], \"bs\", label=\"Iris versicolor\")\n", "plot_svc_decision_boundary(svm_clf1, 4, 5.9)\n", "plt.xlabel(\"Petal length\", fontsize=14)\n", "plt.ylabel(\"Petal width\", fontsize=14)\n", "plt.legend(loc=\"upper left\", fontsize=14)\n", "plt.title(\"$C = {}$\".format(svm_clf1.C), fontsize=16)\n", "plt.axis([4, 5.9, 0.8, 2.8])\n", "\n", "plt.sca(axes[1])\n", "plt.plot(X[:, 0][y==1], X[:, 1][y==1], \"g^\")\n", "plt.plot(X[:, 0][y==0], X[:, 1][y==0], \"bs\")\n", "plot_svc_decision_boundary(svm_clf2, 4, 5.99)\n", "plt.xlabel(\"Petal length\", fontsize=14)\n", "plt.title(\"$C = {}$\".format(svm_clf2.C), fontsize=16)\n", "plt.axis([4, 5.9, 0.8, 2.8])\n", "\n", "save_fig(\"regularization_plot\")" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# Non-linear classification" ] }, { "cell_type": "code", "execution_count": 11, "metadata": {}, "outputs": [], "source": [ "X1D = np.linspace(-4, 4, 9).reshape(-1, 1)\n", "X2D = np.c_[X1D, X1D**2]\n", "y = np.array([0, 0, 1, 1, 1, 1, 1, 0, 0])\n", "\n", "plt.figure(figsize=(10, 3))\n", "\n", "plt.subplot(121)\n", "plt.grid(True, which='both')\n", "plt.axhline(y=0, color='k')\n", "plt.plot(X1D[:, 0][y==0], np.zeros(4), \"bs\")\n", "plt.plot(X1D[:, 0][y==1], np.zeros(5), \"g^\")\n", "plt.gca().get_yaxis().set_ticks([])\n", "plt.xlabel(r\"$x_1$\", fontsize=20)\n", "plt.axis([-4.5, 4.5, -0.2, 0.2])\n", "\n", "plt.subplot(122)\n", "plt.grid(True, which='both')\n", "plt.axhline(y=0, color='k')\n", "plt.axvline(x=0, color='k')\n", "plt.plot(X2D[:, 0][y==0], X2D[:, 1][y==0], \"bs\")\n", "plt.plot(X2D[:, 0][y==1], X2D[:, 1][y==1], \"g^\")\n", "plt.xlabel(r\"$x_1$\", fontsize=20)\n", "plt.ylabel(r\"$x_2$  \", fontsize=20, rotation=0)\n", "plt.gca().get_yaxis().set_ticks([0, 4, 8, 12, 16])\n", "plt.plot([-4.5, 4.5], [6.5, 6.5], \"r--\", linewidth=3)\n", "plt.axis([-4.5, 4.5, -1, 17])\n", "\n", "plt.subplots_adjust(right=1)\n", "\n", "save_fig(\"higher_dimensions_plot\", tight_layout=False)\n", "plt.show()" ] }, { "cell_type": "code", "execution_count": 12, "metadata": {}, "outputs": [], "source": [ "from sklearn.datasets import make_moons\n", "X, y = make_moons(n_samples=100, noise=0.15, random_state=42)\n", "\n", "def plot_dataset(X, y, axes):\n", " plt.plot(X[:, 0][y==0], X[:, 1][y==0], \"bs\")\n", " plt.plot(X[:, 0][y==1], X[:, 1][y==1], \"g^\")\n", " plt.axis(axes)\n", " plt.grid(True, which='both')\n", " plt.xlabel(r\"$x_1$\", fontsize=20)\n", " plt.ylabel(r\"$x_2$\", fontsize=20, rotation=0)\n", "\n", "plot_dataset(X, y, [-1.5, 2.5, -1, 1.5])\n", "plt.show()" ] }, { "cell_type": "code", "execution_count": 13, "metadata": {}, "outputs": [], "source": [ "from sklearn.datasets import make_moons\n", "from sklearn.pipeline import Pipeline\n", "from sklearn.preprocessing import PolynomialFeatures\n", "\n", "polynomial_svm_clf = Pipeline([\n", " (\"poly_features\", PolynomialFeatures(degree=3)),\n", " (\"scaler\", StandardScaler()),\n", " (\"svm_clf\", LinearSVC(C=10, loss=\"hinge\", random_state=42))\n", " ])\n", "\n", "polynomial_svm_clf.fit(X, y)" ] }, { "cell_type": "code", "execution_count": 14, "metadata": {}, "outputs": [], "source": [ "def plot_predictions(clf, axes):\n", " x0s = np.linspace(axes[0], axes[1], 100)\n", " x1s = np.linspace(axes[2], axes[3], 100)\n", " x0, x1 = np.meshgrid(x0s, x1s)\n", " X = np.c_[x0.ravel(), x1.ravel()]\n", " y_pred = clf.predict(X).reshape(x0.shape)\n", " y_decision = clf.decision_function(X).reshape(x0.shape)\n", " plt.contourf(x0, x1, y_pred, cmap=plt.cm.brg, alpha=0.2)\n", " plt.contourf(x0, x1, y_decision, cmap=plt.cm.brg, alpha=0.1)\n", "\n", "plot_predictions(polynomial_svm_clf, [-1.5, 2.5, -1, 1.5])\n", "plot_dataset(X, y, [-1.5, 2.5, -1, 1.5])\n", "\n", "save_fig(\"moons_polynomial_svc_plot\")\n", "plt.show()" ] }, { "cell_type": "code", "execution_count": 15, "metadata": {}, "outputs": [], "source": [ "from sklearn.svm import SVC\n", "\n", "poly_kernel_svm_clf = Pipeline([\n", " (\"scaler\", StandardScaler()),\n", " (\"svm_clf\", SVC(kernel=\"poly\", degree=3, coef0=1, C=5))\n", " ])\n", "poly_kernel_svm_clf.fit(X, y)" ] }, { "cell_type": "code", "execution_count": 16, "metadata": {}, "outputs": [], "source": [ "poly100_kernel_svm_clf = Pipeline([\n", " (\"scaler\", StandardScaler()),\n", " (\"svm_clf\", SVC(kernel=\"poly\", degree=10, coef0=100, C=5))\n", " ])\n", "poly100_kernel_svm_clf.fit(X, y)" ] }, { "cell_type": "code", "execution_count": 17, "metadata": {}, "outputs": [], "source": [ "fig, axes = plt.subplots(ncols=2, figsize=(10.5, 4), sharey=True)\n", "\n", "plt.sca(axes[0])\n", "plot_predictions(poly_kernel_svm_clf, [-1.5, 2.45, -1, 1.5])\n", "plot_dataset(X, y, [-1.5, 2.4, -1, 1.5])\n", "plt.title(r\"$d=3, r=1, C=5$\", fontsize=18)\n", "\n", "plt.sca(axes[1])\n", "plot_predictions(poly100_kernel_svm_clf, [-1.5, 2.45, -1, 1.5])\n", "plot_dataset(X, y, [-1.5, 2.4, -1, 1.5])\n", "plt.title(r\"$d=10, r=100, C=5$\", fontsize=18)\n", "plt.ylabel(\"\")\n", "\n", "save_fig(\"moons_kernelized_polynomial_svc_plot\")\n", "plt.show()" ] }, { "cell_type": "code", "execution_count": 18, "metadata": { "scrolled": true }, "outputs": [], "source": [ "def gaussian_rbf(x, landmark, gamma):\n", " return np.exp(-gamma * np.linalg.norm(x - landmark, axis=1)**2)\n", "\n", "gamma = 0.3\n", "\n", "x1s = np.linspace(-4.5, 4.5, 200).reshape(-1, 1)\n", "x2s = gaussian_rbf(x1s, -2, gamma)\n", "x3s = gaussian_rbf(x1s, 1, gamma)\n", "\n", "XK = np.c_[gaussian_rbf(X1D, -2, gamma), gaussian_rbf(X1D, 1, gamma)]\n", "yk = np.array([0, 0, 1, 1, 1, 1, 1, 0, 0])\n", "\n", "plt.figure(figsize=(10.5, 4))\n", "\n", "plt.subplot(121)\n", "plt.grid(True, which='both')\n", "plt.axhline(y=0, color='k')\n", "plt.scatter(x=[-2, 1], y=[0, 0], s=150, alpha=0.5, c=\"red\")\n", "plt.plot(X1D[:, 0][yk==0], np.zeros(4), \"bs\")\n", "plt.plot(X1D[:, 0][yk==1], np.zeros(5), \"g^\")\n", "plt.plot(x1s, x2s, \"g--\")\n", "plt.plot(x1s, x3s, \"b:\")\n", "plt.gca().get_yaxis().set_ticks([0, 0.25, 0.5, 0.75, 1])\n", "plt.xlabel(r\"$x_1$\", fontsize=20)\n", "plt.ylabel(r\"Similarity\", fontsize=14)\n", "plt.annotate(r'$\\mathbf{x}$',\n", " xy=(X1D[3, 0], 0),\n", " xytext=(-0.5, 0.20),\n", " ha=\"center\",\n", " arrowprops=dict(facecolor='black', shrink=0.1),\n", " fontsize=18,\n", " )\n", "plt.text(-2, 0.9, \"$x_2$\", ha=\"center\", fontsize=20)\n", "plt.text(1, 0.9, \"$x_3$\", ha=\"center\", fontsize=20)\n", "plt.axis([-4.5, 4.5, -0.1, 1.1])\n", "\n", "plt.subplot(122)\n", "plt.grid(True, which='both')\n", "plt.axhline(y=0, color='k')\n", "plt.axvline(x=0, color='k')\n", "plt.plot(XK[:, 0][yk==0], XK[:, 1][yk==0], \"bs\")\n", "plt.plot(XK[:, 0][yk==1], XK[:, 1][yk==1], \"g^\")\n", "plt.xlabel(r\"$x_2$\", fontsize=20)\n", "plt.ylabel(r\"$x_3$  \", fontsize=20, rotation=0)\n", "plt.annotate(r'$\\phi\\left(\\mathbf{x}\\right)$',\n", " xy=(XK[3, 0], XK[3, 1]),\n", " xytext=(0.65, 0.50),\n", " ha=\"center\",\n", " arrowprops=dict(facecolor='black', shrink=0.1),\n", " fontsize=18,\n", " )\n", "plt.plot([-0.1, 1.1], [0.57, -0.1], \"r--\", linewidth=3)\n", "plt.axis([-0.1, 1.1, -0.1, 1.1])\n", " \n", "plt.subplots_adjust(right=1)\n", "\n", "save_fig(\"kernel_method_plot\")\n", "plt.show()" ] }, { "cell_type": "code", "execution_count": 19, "metadata": {}, "outputs": [], "source": [ "x1_example = X1D[3, 0]\n", "for landmark in (-2, 1):\n", " k = gaussian_rbf(np.array([[x1_example]]), np.array([[landmark]]), gamma)\n", " print(\"Phi({}, {}) = {}\".format(x1_example, landmark, k))" ] }, { "cell_type": "code", "execution_count": 20, "metadata": {}, "outputs": [], "source": [ "rbf_kernel_svm_clf = Pipeline([\n", " (\"scaler\", StandardScaler()),\n", " (\"svm_clf\", SVC(kernel=\"rbf\", gamma=5, C=0.001))\n", " ])\n", "rbf_kernel_svm_clf.fit(X, y)" ] }, { "cell_type": "code", "execution_count": 21, "metadata": { "scrolled": true }, "outputs": [], "source": [ "from sklearn.svm import SVC\n", "\n", "gamma1, gamma2 = 0.1, 5\n", "C1, C2 = 0.001, 1000\n", "hyperparams = (gamma1, C1), (gamma1, C2), (gamma2, C1), (gamma2, C2)\n", "\n", "svm_clfs = []\n", "for gamma, C in hyperparams:\n", " rbf_kernel_svm_clf = Pipeline([\n", " (\"scaler\", StandardScaler()),\n", " (\"svm_clf\", SVC(kernel=\"rbf\", gamma=gamma, C=C))\n", " ])\n", " rbf_kernel_svm_clf.fit(X, y)\n", " svm_clfs.append(rbf_kernel_svm_clf)\n", "\n", "fig, axes = plt.subplots(nrows=2, ncols=2, figsize=(10.5, 7), sharex=True, sharey=True)\n", "\n", "for i, svm_clf in enumerate(svm_clfs):\n", " plt.sca(axes[i // 2, i % 2])\n", " plot_predictions(svm_clf, [-1.5, 2.45, -1, 1.5])\n", " plot_dataset(X, y, [-1.5, 2.45, -1, 1.5])\n", " gamma, C = hyperparams[i]\n", " plt.title(r\"$\\gamma = {}, C = {}$\".format(gamma, C), fontsize=16)\n", " if i in (0, 1):\n", " plt.xlabel(\"\")\n", " if i in (1, 3):\n", " plt.ylabel(\"\")\n", "\n", "save_fig(\"moons_rbf_svc_plot\")\n", "plt.show()" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# Regression\n" ] }, { "cell_type": "code", "execution_count": 22, "metadata": {}, "outputs": [], "source": [ "np.random.seed(42)\n", "m = 50\n", "X = 2 * np.random.rand(m, 1)\n", "y = (4 + 3 * X + np.random.randn(m, 1)).ravel()" ] }, { "cell_type": "code", "execution_count": 23, "metadata": {}, "outputs": [], "source": [ "from sklearn.svm import LinearSVR\n", "\n", "svm_reg = LinearSVR(epsilon=1.5, random_state=42)\n", "svm_reg.fit(X, y)" ] }, { "cell_type": "code", "execution_count": 24, "metadata": {}, "outputs": [], "source": [ "svm_reg1 = LinearSVR(epsilon=1.5, random_state=42)\n", "svm_reg2 = LinearSVR(epsilon=0.5, random_state=42)\n", "svm_reg1.fit(X, y)\n", "svm_reg2.fit(X, y)\n", "\n", "def find_support_vectors(svm_reg, X, y):\n", " y_pred = svm_reg.predict(X)\n", " off_margin = (np.abs(y - y_pred) >= svm_reg.epsilon)\n", " return np.argwhere(off_margin)\n", "\n", "svm_reg1.support_ = find_support_vectors(svm_reg1, X, y)\n", "svm_reg2.support_ = find_support_vectors(svm_reg2, X, y)\n", "\n", "eps_x1 = 1\n", "eps_y_pred = svm_reg1.predict([[eps_x1]])" ] }, { "cell_type": "code", "execution_count": 25, "metadata": {}, "outputs": [], "source": [ "def plot_svm_regression(svm_reg, X, y, axes):\n", " x1s = np.linspace(axes[0], axes[1], 100).reshape(100, 1)\n", " y_pred = svm_reg.predict(x1s)\n", " plt.plot(x1s, y_pred, \"k-\", linewidth=2, label=r\"$\\hat{y}$\")\n", " plt.plot(x1s, y_pred + svm_reg.epsilon, \"k--\")\n", " plt.plot(x1s, y_pred - svm_reg.epsilon, \"k--\")\n", " plt.scatter(X[svm_reg.support_], y[svm_reg.support_], s=180, facecolors='#FFAAAA')\n", " plt.plot(X, y, \"bo\")\n", " plt.xlabel(r\"$x_1$\", fontsize=18)\n", " plt.legend(loc=\"upper left\", fontsize=18)\n", " plt.axis(axes)\n", "\n", "fig, axes = plt.subplots(ncols=2, figsize=(9, 4), sharey=True)\n", "plt.sca(axes[0])\n", "plot_svm_regression(svm_reg1, X, y, [0, 2, 3, 11])\n", "plt.title(r\"$\\epsilon = {}$\".format(svm_reg1.epsilon), fontsize=18)\n", "plt.ylabel(r\"$y$\", fontsize=18, rotation=0)\n", "#plt.plot([eps_x1, eps_x1], [eps_y_pred, eps_y_pred - svm_reg1.epsilon], \"k-\", linewidth=2)\n", "plt.annotate(\n", " '', xy=(eps_x1, eps_y_pred), xycoords='data',\n", " xytext=(eps_x1, eps_y_pred - svm_reg1.epsilon),\n", " textcoords='data', arrowprops={'arrowstyle': '<->', 'linewidth': 1.5}\n", " )\n", "plt.text(0.91, 5.6, r\"$\\epsilon$\", fontsize=20)\n", "plt.sca(axes[1])\n", "plot_svm_regression(svm_reg2, X, y, [0, 2, 3, 11])\n", "plt.title(r\"$\\epsilon = {}$\".format(svm_reg2.epsilon), fontsize=18)\n", "save_fig(\"svm_regression_plot\")\n", "plt.show()" ] }, { "cell_type": "code", "execution_count": 26, "metadata": {}, "outputs": [], "source": [ "np.random.seed(42)\n", "m = 100\n", "X = 2 * np.random.rand(m, 1) - 1\n", "y = (0.2 + 0.1 * X + 0.5 * X**2 + np.random.randn(m, 1)/10).ravel()" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "**Note**: to be future-proof, we set `gamma=\"scale\"`, as this will be the default value in Scikit-Learn 0.22." ] }, { "cell_type": "code", "execution_count": 27, "metadata": {}, "outputs": [], "source": [ "from sklearn.svm import SVR\n", "\n", "svm_poly_reg = SVR(kernel=\"poly\", degree=2, C=100, epsilon=0.1, gamma=\"scale\")\n", "svm_poly_reg.fit(X, y)" ] }, { "cell_type": "code", "execution_count": 28, "metadata": {}, "outputs": [], "source": [ "from sklearn.svm import SVR\n", "\n", "svm_poly_reg1 = SVR(kernel=\"poly\", degree=2, C=100, epsilon=0.1, gamma=\"scale\")\n", "svm_poly_reg2 = SVR(kernel=\"poly\", degree=2, C=0.01, epsilon=0.1, gamma=\"scale\")\n", "svm_poly_reg1.fit(X, y)\n", "svm_poly_reg2.fit(X, y)" ] }, { "cell_type": "code", "execution_count": 29, "metadata": {}, "outputs": [], "source": [ "fig, axes = plt.subplots(ncols=2, figsize=(9, 4), sharey=True)\n", "plt.sca(axes[0])\n", "plot_svm_regression(svm_poly_reg1, X, y, [-1, 1, 0, 1])\n", "plt.title(r\"$degree={}, C={}, \\epsilon = {}$\".format(svm_poly_reg1.degree, svm_poly_reg1.C, svm_poly_reg1.epsilon), fontsize=18)\n", "plt.ylabel(r\"$y$\", fontsize=18, rotation=0)\n", "plt.sca(axes[1])\n", "plot_svm_regression(svm_poly_reg2, X, y, [-1, 1, 0, 1])\n", "plt.title(r\"$degree={}, C={}, \\epsilon = {}$\".format(svm_poly_reg2.degree, svm_poly_reg2.C, svm_poly_reg2.epsilon), fontsize=18)\n", "save_fig(\"svm_with_polynomial_kernel_plot\")\n", "plt.show()" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# Under the hood" ] }, { "cell_type": "code", "execution_count": 30, "metadata": {}, "outputs": [], "source": [ "iris = datasets.load_iris()\n", "X = iris[\"data\"][:, (2, 3)] # petal length, petal width\n", "y = (iris[\"target\"] == 2).astype(np.float64) # Iris virginica" ] }, { "cell_type": "code", "execution_count": 31, "metadata": {}, "outputs": [], "source": [ "from mpl_toolkits.mplot3d import Axes3D\n", "\n", "def plot_3D_decision_function(ax, w, b, x1_lim=[4, 6], x2_lim=[0.8, 2.8]):\n", " x1_in_bounds = (X[:, 0] > x1_lim[0]) & (X[:, 0] < x1_lim[1])\n", " X_crop = X[x1_in_bounds]\n", " y_crop = y[x1_in_bounds]\n", " x1s = np.linspace(x1_lim[0], x1_lim[1], 20)\n", " x2s = np.linspace(x2_lim[0], x2_lim[1], 20)\n", " x1, x2 = np.meshgrid(x1s, x2s)\n", " xs = np.c_[x1.ravel(), x2.ravel()]\n", " df = (xs.dot(w) + b).reshape(x1.shape)\n", " m = 1 / np.linalg.norm(w)\n", " boundary_x2s = -x1s*(w[0]/w[1])-b/w[1]\n", " margin_x2s_1 = -x1s*(w[0]/w[1])-(b-1)/w[1]\n", " margin_x2s_2 = -x1s*(w[0]/w[1])-(b+1)/w[1]\n", " ax.plot_surface(x1s, x2, np.zeros_like(x1),\n", " color=\"b\", alpha=0.2, cstride=100, rstride=100)\n", " ax.plot(x1s, boundary_x2s, 0, \"k-\", linewidth=2, label=r\"$h=0$\")\n", " ax.plot(x1s, margin_x2s_1, 0, \"k--\", linewidth=2, label=r\"$h=\\pm 1$\")\n", " ax.plot(x1s, margin_x2s_2, 0, \"k--\", linewidth=2)\n", " ax.plot(X_crop[:, 0][y_crop==1], X_crop[:, 1][y_crop==1], 0, \"g^\")\n", " ax.plot_wireframe(x1, x2, df, alpha=0.3, color=\"k\")\n", " ax.plot(X_crop[:, 0][y_crop==0], X_crop[:, 1][y_crop==0], 0, \"bs\")\n", " ax.axis(x1_lim + x2_lim)\n", " ax.text(4.5, 2.5, 3.8, \"Decision function $h$\", fontsize=16)\n", " ax.set_xlabel(r\"Petal length\", fontsize=16, labelpad=10)\n", " ax.set_ylabel(r\"Petal width\", fontsize=16, labelpad=10)\n", " ax.set_zlabel(r\"$h = \\mathbf{w}^T \\mathbf{x} + b$\", fontsize=18, labelpad=5)\n", " ax.legend(loc=\"upper left\", fontsize=16)\n", "\n", "fig = plt.figure(figsize=(11, 6))\n", "ax1 = fig.add_subplot(111, projection='3d')\n", "plot_3D_decision_function(ax1, w=svm_clf2.coef_[0], b=svm_clf2.intercept_[0])\n", "\n", "save_fig(\"iris_3D_plot\")\n", "plt.show()" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# Small weight vector results in a large margin" ] }, { "cell_type": "code", "execution_count": 32, "metadata": {}, "outputs": [], "source": [ "def plot_2D_decision_function(w, b, ylabel=True, x1_lim=[-3, 3]):\n", " x1 = np.linspace(x1_lim[0], x1_lim[1], 200)\n", " y = w * x1 + b\n", " m = 1 / w\n", "\n", " plt.plot(x1, y)\n", " plt.plot(x1_lim, [1, 1], \"k:\")\n", " plt.plot(x1_lim, [-1, -1], \"k:\")\n", " plt.axhline(y=0, color='k')\n", " plt.axvline(x=0, color='k')\n", " plt.plot([m, m], [0, 1], \"k--\")\n", " plt.plot([-m, -m], [0, -1], \"k--\")\n", " plt.plot([-m, m], [0, 0], \"k-o\", linewidth=3)\n", " plt.axis(x1_lim + [-2, 2])\n", " plt.xlabel(r\"$x_1$\", fontsize=16)\n", " if ylabel:\n", " plt.ylabel(r\"$w_1 x_1$  \", rotation=0, fontsize=16)\n", " plt.title(r\"$w_1 = {}$\".format(w), fontsize=16)\n", "\n", "fig, axes = plt.subplots(ncols=2, figsize=(9, 3.2), sharey=True)\n", "plt.sca(axes[0])\n", "plot_2D_decision_function(1, 0)\n", "plt.sca(axes[1])\n", "plot_2D_decision_function(0.5, 0, ylabel=False)\n", "save_fig(\"small_w_large_margin_plot\")\n", "plt.show()" ] }, { "cell_type": "code", "execution_count": 33, "metadata": {}, "outputs": [], "source": [ "from sklearn.svm import SVC\n", "from sklearn import datasets\n", "\n", "iris = datasets.load_iris()\n", "X = iris[\"data\"][:, (2, 3)] # petal length, petal width\n", "y = (iris[\"target\"] == 2).astype(np.float64) # Iris virginica\n", "\n", "svm_clf = SVC(kernel=\"linear\", C=1)\n", "svm_clf.fit(X, y)\n", "svm_clf.predict([[5.3, 1.3]])" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# Hinge loss" ] }, { "cell_type": "code", "execution_count": 34, "metadata": {}, "outputs": [], "source": [ "t = np.linspace(-2, 4, 200)\n", "h = np.where(1 - t < 0, 0, 1 - t) # max(0, 1-t)\n", "\n", "plt.figure(figsize=(5,2.8))\n", "plt.plot(t, h, \"b-\", linewidth=2, label=\"$max(0, 1 - t)$\")\n", "plt.grid(True, which='both')\n", "plt.axhline(y=0, color='k')\n", "plt.axvline(x=0, color='k')\n", "plt.yticks(np.arange(-1, 2.5, 1))\n", "plt.xlabel(\"$t$\", fontsize=16)\n", "plt.axis([-2, 4, -1, 2.5])\n", "plt.legend(loc=\"upper right\", fontsize=16)\n", "save_fig(\"hinge_plot\")\n", "plt.show()" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# Extra material" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Training time" ] }, { "cell_type": "code", "execution_count": 35, "metadata": {}, "outputs": [], "source": [ "X, y = make_moons(n_samples=1000, noise=0.4, random_state=42)\n", "plt.plot(X[:, 0][y==0], X[:, 1][y==0], \"bs\")\n", "plt.plot(X[:, 0][y==1], X[:, 1][y==1], \"g^\")" ] }, { "cell_type": "code", "execution_count": 36, "metadata": {}, "outputs": [], "source": [ "import time\n", "\n", "tol = 0.1\n", "tols = []\n", "times = []\n", "for i in range(10):\n", " svm_clf = SVC(kernel=\"poly\", gamma=3, C=10, tol=tol, verbose=1)\n", " t1 = time.time()\n", " svm_clf.fit(X, y)\n", " t2 = time.time()\n", " times.append(t2-t1)\n", " tols.append(tol)\n", " print(i, tol, t2-t1)\n", " tol /= 10\n", "plt.semilogx(tols, times, \"bo-\")\n", "plt.xlabel(\"Tolerance\", fontsize=16)\n", "plt.ylabel(\"Time (seconds)\", fontsize=16)\n", "plt.grid(True)\n", "plt.show()" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Linear SVM classifier implementation using Batch Gradient Descent" ] }, { "cell_type": "code", "execution_count": 37, "metadata": {}, "outputs": [], "source": [ "# Training set\n", "X = iris[\"data\"][:, (2, 3)] # petal length, petal width\n", "y = (iris[\"target\"] == 2).astype(np.float64).reshape(-1, 1) # Iris virginica" ] }, { "cell_type": "code", "execution_count": 38, "metadata": {}, "outputs": [], "source": [ "from sklearn.base import BaseEstimator\n", "\n", "class MyLinearSVC(BaseEstimator):\n", " def __init__(self, C=1, eta0=1, eta_d=10000, n_epochs=1000, random_state=None):\n", " self.C = C\n", " self.eta0 = eta0\n", " self.n_epochs = n_epochs\n", " self.random_state = random_state\n", " self.eta_d = eta_d\n", "\n", " def eta(self, epoch):\n", " return self.eta0 / (epoch + self.eta_d)\n", " \n", " def fit(self, X, y):\n", " # Random initialization\n", " if self.random_state:\n", " np.random.seed(self.random_state)\n", " w = np.random.randn(X.shape[1], 1) # n feature weights\n", " b = 0\n", "\n", " m = len(X)\n", " t = y * 2 - 1 # -1 if t==0, +1 if t==1\n", " X_t = X * t\n", " self.Js=[]\n", "\n", " # Training\n", " for epoch in range(self.n_epochs):\n", " support_vectors_idx = (X_t.dot(w) + t * b < 1).ravel()\n", " X_t_sv = X_t[support_vectors_idx]\n", " t_sv = t[support_vectors_idx]\n", "\n", " J = 1/2 * np.sum(w * w) + self.C * (np.sum(1 - X_t_sv.dot(w)) - b * np.sum(t_sv))\n", " self.Js.append(J)\n", "\n", " w_gradient_vector = w - self.C * np.sum(X_t_sv, axis=0).reshape(-1, 1)\n", " b_derivative = -self.C * np.sum(t_sv)\n", " \n", " w = w - self.eta(epoch) * w_gradient_vector\n", " b = b - self.eta(epoch) * b_derivative\n", " \n", "\n", " self.intercept_ = np.array([b])\n", " self.coef_ = np.array([w])\n", " support_vectors_idx = (X_t.dot(w) + t * b < 1).ravel()\n", " self.support_vectors_ = X[support_vectors_idx]\n", " return self\n", "\n", " def decision_function(self, X):\n", " return X.dot(self.coef_[0]) + self.intercept_[0]\n", "\n", " def predict(self, X):\n", " return (self.decision_function(X) >= 0).astype(np.float64)\n", "\n", "C=2\n", "svm_clf = MyLinearSVC(C=C, eta0 = 10, eta_d = 1000, n_epochs=60000, random_state=2)\n", "svm_clf.fit(X, y)\n", "svm_clf.predict(np.array([[5, 2], [4, 1]]))" ] }, { "cell_type": "code", "execution_count": 39, "metadata": {}, "outputs": [], "source": [ "plt.plot(range(svm_clf.n_epochs), svm_clf.Js)\n", "plt.axis([0, svm_clf.n_epochs, 0, 100])" ] }, { "cell_type": "code", "execution_count": 40, "metadata": {}, "outputs": [], "source": [ "print(svm_clf.intercept_, svm_clf.coef_)" ] }, { "cell_type": "code", "execution_count": 41, "metadata": {}, "outputs": [], "source": [ "svm_clf2 = SVC(kernel=\"linear\", C=C)\n", "svm_clf2.fit(X, y.ravel())\n", "print(svm_clf2.intercept_, svm_clf2.coef_)" ] }, { "cell_type": "code", "execution_count": 42, "metadata": {}, "outputs": [], "source": [ "yr = y.ravel()\n", "fig, axes = plt.subplots(ncols=2, figsize=(11, 3.2), sharey=True)\n", "plt.sca(axes[0])\n", "plt.plot(X[:, 0][yr==1], X[:, 1][yr==1], \"g^\", label=\"Iris virginica\")\n", "plt.plot(X[:, 0][yr==0], X[:, 1][yr==0], \"bs\", label=\"Not Iris virginica\")\n", "plot_svc_decision_boundary(svm_clf, 4, 6)\n", "plt.xlabel(\"Petal length\", fontsize=14)\n", "plt.ylabel(\"Petal width\", fontsize=14)\n", "plt.title(\"MyLinearSVC\", fontsize=14)\n", "plt.axis([4, 6, 0.8, 2.8])\n", "plt.legend(loc=\"upper left\")\n", "\n", "plt.sca(axes[1])\n", "plt.plot(X[:, 0][yr==1], X[:, 1][yr==1], \"g^\")\n", "plt.plot(X[:, 0][yr==0], X[:, 1][yr==0], \"bs\")\n", "plot_svc_decision_boundary(svm_clf2, 4, 6)\n", "plt.xlabel(\"Petal length\", fontsize=14)\n", "plt.title(\"SVC\", fontsize=14)\n", "plt.axis([4, 6, 0.8, 2.8])\n" ] }, { "cell_type": "code", "execution_count": 43, "metadata": { "scrolled": true }, "outputs": [], "source": [ "from sklearn.linear_model import SGDClassifier\n", "\n", "sgd_clf = SGDClassifier(loss=\"hinge\", alpha=0.017, max_iter=1000, tol=1e-3, random_state=42)\n", "sgd_clf.fit(X, y.ravel())\n", "\n", "m = len(X)\n", "t = y * 2 - 1 # -1 if t==0, +1 if t==1\n", "X_b = np.c_[np.ones((m, 1)), X] # Add bias input x0=1\n", "X_b_t = X_b * t\n", "sgd_theta = np.r_[sgd_clf.intercept_[0], sgd_clf.coef_[0]]\n", "print(sgd_theta)\n", "support_vectors_idx = (X_b_t.dot(sgd_theta) < 1).ravel()\n", "sgd_clf.support_vectors_ = X[support_vectors_idx]\n", "sgd_clf.C = C\n", "\n", "plt.figure(figsize=(5.5,3.2))\n", "plt.plot(X[:, 0][yr==1], X[:, 1][yr==1], \"g^\")\n", "plt.plot(X[:, 0][yr==0], X[:, 1][yr==0], \"bs\")\n", "plot_svc_decision_boundary(sgd_clf, 4, 6)\n", "plt.xlabel(\"Petal length\", fontsize=14)\n", "plt.ylabel(\"Petal width\", fontsize=14)\n", "plt.title(\"SGDClassifier\", fontsize=14)\n", "plt.axis([4, 6, 0.8, 2.8])\n" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# Exercise solutions" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 1. to 7." ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "See appendix A." ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# 8." ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "_Exercise: train a `LinearSVC` on a linearly separable dataset. Then train an `SVC` and a `SGDClassifier` on the same dataset. See if you can get them to produce roughly the same model._" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Let's use the Iris dataset: the Iris Setosa and Iris Versicolor classes are linearly separable." ] }, { "cell_type": "code", "execution_count": 44, "metadata": {}, "outputs": [], "source": [ "from sklearn import datasets\n", "\n", "iris = datasets.load_iris()\n", "X = iris[\"data\"][:, (2, 3)] # petal length, petal width\n", "y = iris[\"target\"]\n", "\n", "setosa_or_versicolor = (y == 0) | (y == 1)\n", "X = X[setosa_or_versicolor]\n", "y = y[setosa_or_versicolor]" ] }, { "cell_type": "code", "execution_count": 45, "metadata": {}, "outputs": [], "source": [ "from sklearn.svm import SVC, LinearSVC\n", "from sklearn.linear_model import SGDClassifier\n", "from sklearn.preprocessing import StandardScaler\n", "\n", "C = 5\n", "alpha = 1 / (C * len(X))\n", "\n", "lin_clf = LinearSVC(loss=\"hinge\", C=C, random_state=42)\n", "svm_clf = SVC(kernel=\"linear\", C=C)\n", "sgd_clf = SGDClassifier(loss=\"hinge\", learning_rate=\"constant\", eta0=0.001, alpha=alpha,\n", " max_iter=1000, tol=1e-3, random_state=42)\n", "\n", "scaler = StandardScaler()\n", "X_scaled = scaler.fit_transform(X)\n", "\n", "lin_clf.fit(X_scaled, y)\n", "svm_clf.fit(X_scaled, y)\n", "sgd_clf.fit(X_scaled, y)\n", "\n", "print(\"LinearSVC: \", lin_clf.intercept_, lin_clf.coef_)\n", "print(\"SVC: \", svm_clf.intercept_, svm_clf.coef_)\n", "print(\"SGDClassifier(alpha={:.5f}):\".format(sgd_clf.alpha), sgd_clf.intercept_, sgd_clf.coef_)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Let's plot the decision boundaries of these three models:" ] }, { "cell_type": "code", "execution_count": 46, "metadata": {}, "outputs": [], "source": [ "# Compute the slope and bias of each decision boundary\n", "w1 = -lin_clf.coef_[0, 0]/lin_clf.coef_[0, 1]\n", "b1 = -lin_clf.intercept_[0]/lin_clf.coef_[0, 1]\n", "w2 = -svm_clf.coef_[0, 0]/svm_clf.coef_[0, 1]\n", "b2 = -svm_clf.intercept_[0]/svm_clf.coef_[0, 1]\n", "w3 = -sgd_clf.coef_[0, 0]/sgd_clf.coef_[0, 1]\n", "b3 = -sgd_clf.intercept_[0]/sgd_clf.coef_[0, 1]\n", "\n", "# Transform the decision boundary lines back to the original scale\n", "line1 = scaler.inverse_transform([[-10, -10 * w1 + b1], [10, 10 * w1 + b1]])\n", "line2 = scaler.inverse_transform([[-10, -10 * w2 + b2], [10, 10 * w2 + b2]])\n", "line3 = scaler.inverse_transform([[-10, -10 * w3 + b3], [10, 10 * w3 + b3]])\n", "\n", "# Plot all three decision boundaries\n", "plt.figure(figsize=(11, 4))\n", "plt.plot(line1[:, 0], line1[:, 1], \"k:\", label=\"LinearSVC\")\n", "plt.plot(line2[:, 0], line2[:, 1], \"b--\", linewidth=2, label=\"SVC\")\n", "plt.plot(line3[:, 0], line3[:, 1], \"r-\", label=\"SGDClassifier\")\n", "plt.plot(X[:, 0][y==1], X[:, 1][y==1], \"bs\") # label=\"Iris versicolor\"\n", "plt.plot(X[:, 0][y==0], X[:, 1][y==0], \"yo\") # label=\"Iris setosa\"\n", "plt.xlabel(\"Petal length\", fontsize=14)\n", "plt.ylabel(\"Petal width\", fontsize=14)\n", "plt.legend(loc=\"upper center\", fontsize=14)\n", "plt.axis([0, 5.5, 0, 2])\n", "\n", "plt.show()" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Close enough!" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# 9." ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "_Exercise: train an SVM classifier on the MNIST dataset. Since SVM classifiers are binary classifiers, you will need to use one-versus-all to classify all 10 digits. You may want to tune the hyperparameters using small validation sets to speed up the process. What accuracy can you reach?_" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "First, let's load the dataset and split it into a training set and a test set. We could use `train_test_split()` but people usually just take the first 60,000 instances for the training set, and the last 10,000 instances for the test set (this makes it possible to compare your model's performance with others): " ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "**Warning:** since Scikit-Learn 0.24, `fetch_openml()` returns a Pandas `DataFrame` by default. To avoid this, we use `as_frame=False`." ] }, { "cell_type": "code", "execution_count": 47, "metadata": {}, "outputs": [], "source": [ "from sklearn.datasets import fetch_openml\n", "mnist = fetch_openml('mnist_784', version=1, cache=True, as_frame=False)\n", "\n", "X = mnist[\"data\"]\n", "y = mnist[\"target\"].astype(np.uint8)\n", "\n", "X_train = X[:60000]\n", "y_train = y[:60000]\n", "X_test = X[60000:]\n", "y_test = y[60000:]" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Many training algorithms are sensitive to the order of the training instances, so it's generally good practice to shuffle them first. However, the dataset is already shuffled, so we do not need to do it." ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Let's start simple, with a linear SVM classifier. It will automatically use the One-vs-All (also called One-vs-the-Rest, OvR) strategy, so there's nothing special we need to do. Easy!\n", "\n", "**Warning**: this may take a few minutes depending on your hardware." ] }, { "cell_type": "code", "execution_count": 48, "metadata": {}, "outputs": [], "source": [ "lin_clf = LinearSVC(random_state=42)\n", "lin_clf.fit(X_train, y_train)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Let's make predictions on the training set and measure the accuracy (we don't want to measure it on the test set yet, since we have not selected and trained the final model yet):" ] }, { "cell_type": "code", "execution_count": 49, "metadata": {}, "outputs": [], "source": [ "from sklearn.metrics import accuracy_score\n", "\n", "y_pred = lin_clf.predict(X_train)\n", "accuracy_score(y_train, y_pred)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Okay, 89.5% accuracy on MNIST is pretty bad. This linear model is certainly too simple for MNIST, but perhaps we just needed to scale the data first:" ] }, { "cell_type": "code", "execution_count": 50, "metadata": {}, "outputs": [], "source": [ "scaler = StandardScaler()\n", "X_train_scaled = scaler.fit_transform(X_train.astype(np.float32))\n", "X_test_scaled = scaler.transform(X_test.astype(np.float32))" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "**Warning**: this may take a few minutes depending on your hardware." ] }, { "cell_type": "code", "execution_count": 51, "metadata": {}, "outputs": [], "source": [ "lin_clf = LinearSVC(random_state=42)\n", "lin_clf.fit(X_train_scaled, y_train)" ] }, { "cell_type": "code", "execution_count": 52, "metadata": {}, "outputs": [], "source": [ "y_pred = lin_clf.predict(X_train_scaled)\n", "accuracy_score(y_train, y_pred)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "That's much better (we cut the error rate by about 25%), but still not great at all for MNIST. If we want to use an SVM, we will have to use a kernel. Let's try an `SVC` with an RBF kernel (the default)." ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "**Note**: to be future-proof we set `gamma=\"scale\"` since it will be the default value in Scikit-Learn 0.22." ] }, { "cell_type": "code", "execution_count": 53, "metadata": {}, "outputs": [], "source": [ "svm_clf = SVC(gamma=\"scale\")\n", "svm_clf.fit(X_train_scaled[:10000], y_train[:10000])" ] }, { "cell_type": "code", "execution_count": 54, "metadata": {}, "outputs": [], "source": [ "y_pred = svm_clf.predict(X_train_scaled)\n", "accuracy_score(y_train, y_pred)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "That's promising, we get better performance even though we trained the model on 6 times less data. Let's tune the hyperparameters by doing a randomized search with cross validation. We will do this on a small dataset just to speed up the process:" ] }, { "cell_type": "code", "execution_count": 55, "metadata": {}, "outputs": [], "source": [ "from sklearn.model_selection import RandomizedSearchCV\n", "from scipy.stats import reciprocal, uniform\n", "\n", "param_distributions = {\"gamma\": reciprocal(0.001, 0.1), \"C\": uniform(1, 10)}\n", "rnd_search_cv = RandomizedSearchCV(svm_clf, param_distributions, n_iter=10, verbose=2, cv=3)\n", "rnd_search_cv.fit(X_train_scaled[:1000], y_train[:1000])" ] }, { "cell_type": "code", "execution_count": 56, "metadata": {}, "outputs": [], "source": [ "rnd_search_cv.best_estimator_" ] }, { "cell_type": "code", "execution_count": 57, "metadata": {}, "outputs": [], "source": [ "rnd_search_cv.best_score_" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "This looks pretty low but remember we only trained the model on 1,000 instances. Let's retrain the best estimator on the whole training set:" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "**Warning**: the following cell may take hours to run, depending on your hardware." ] }, { "cell_type": "code", "execution_count": 58, "metadata": {}, "outputs": [], "source": [ "rnd_search_cv.best_estimator_.fit(X_train_scaled, y_train)" ] }, { "cell_type": "code", "execution_count": 59, "metadata": {}, "outputs": [], "source": [ "y_pred = rnd_search_cv.best_estimator_.predict(X_train_scaled)\n", "accuracy_score(y_train, y_pred)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Ah, this looks good! Let's select this model. Now we can test it on the test set:" ] }, { "cell_type": "code", "execution_count": 60, "metadata": {}, "outputs": [], "source": [ "y_pred = rnd_search_cv.best_estimator_.predict(X_test_scaled)\n", "accuracy_score(y_test, y_pred)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Not too bad, but apparently the model is overfitting slightly. It's tempting to tweak the hyperparameters a bit more (e.g. decreasing `C` and/or `gamma`), but we would run the risk of overfitting the test set. Other people have found that the hyperparameters `C=5` and `gamma=0.005` yield even better performance (over 98% accuracy). By running the randomized search for longer and on a larger part of the training set, you may be able to find this as well." ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 10." ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "_Exercise: train an SVM regressor on the California housing dataset._" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Let's load the dataset using Scikit-Learn's `fetch_california_housing()` function:" ] }, { "cell_type": "code", "execution_count": 61, "metadata": {}, "outputs": [], "source": [ "from sklearn.datasets import fetch_california_housing\n", "\n", "housing = fetch_california_housing()\n", "X = housing[\"data\"]\n", "y = housing[\"target\"]" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Split it into a training set and a test set:" ] }, { "cell_type": "code", "execution_count": 62, "metadata": {}, "outputs": [], "source": [ "from sklearn.model_selection import train_test_split\n", "\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": {}, "source": [ "Don't forget to scale the data:" ] }, { "cell_type": "code", "execution_count": 63, "metadata": {}, "outputs": [], "source": [ "from sklearn.preprocessing import StandardScaler\n", "\n", "scaler = StandardScaler()\n", "X_train_scaled = scaler.fit_transform(X_train)\n", "X_test_scaled = scaler.transform(X_test)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Let's train a simple `LinearSVR` first:" ] }, { "cell_type": "code", "execution_count": 64, "metadata": {}, "outputs": [], "source": [ "from sklearn.svm import LinearSVR\n", "\n", "lin_svr = LinearSVR(random_state=42)\n", "lin_svr.fit(X_train_scaled, y_train)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Let's see how it performs on the training set:" ] }, { "cell_type": "code", "execution_count": 65, "metadata": {}, "outputs": [], "source": [ "from sklearn.metrics import mean_squared_error\n", "\n", "y_pred = lin_svr.predict(X_train_scaled)\n", "mse = mean_squared_error(y_train, y_pred)\n", "mse" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Let's look at the RMSE:" ] }, { "cell_type": "code", "execution_count": 66, "metadata": {}, "outputs": [], "source": [ "np.sqrt(mse)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "In this training set, the targets are tens of thousands of dollars. The RMSE gives a rough idea of the kind of error you should expect (with a higher weight for large errors): so with this model we can expect errors somewhere around $10,000. Not great. Let's see if we can do better with an RBF Kernel. We will use randomized search with cross validation to find the appropriate hyperparameter values for `C` and `gamma`:" ] }, { "cell_type": "code", "execution_count": 67, "metadata": {}, "outputs": [], "source": [ "from sklearn.svm import SVR\n", "from sklearn.model_selection import RandomizedSearchCV\n", "from scipy.stats import reciprocal, uniform\n", "\n", "param_distributions = {\"gamma\": reciprocal(0.001, 0.1), \"C\": uniform(1, 10)}\n", "rnd_search_cv = RandomizedSearchCV(SVR(), param_distributions, n_iter=10, verbose=2, cv=3, random_state=42)\n", "rnd_search_cv.fit(X_train_scaled, y_train)" ] }, { "cell_type": "code", "execution_count": 68, "metadata": {}, "outputs": [], "source": [ "rnd_search_cv.best_estimator_" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Now let's measure the RMSE on the training set:" ] }, { "cell_type": "code", "execution_count": 69, "metadata": {}, "outputs": [], "source": [ "y_pred = rnd_search_cv.best_estimator_.predict(X_train_scaled)\n", "mse = mean_squared_error(y_train, y_pred)\n", "np.sqrt(mse)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Looks much better than the linear model. Let's select this model and evaluate it on the test set:" ] }, { "cell_type": "code", "execution_count": 70, "metadata": {}, "outputs": [], "source": [ "y_pred = rnd_search_cv.best_estimator_.predict(X_test_scaled)\n", "mse = mean_squared_error(y_test, y_pred)\n", "np.sqrt(mse)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [] } ], "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": "3.7.9" }, "nav_menu": {}, "toc": { "navigate_menu": true, "number_sections": true, "sideBar": true, "threshold": 6, "toc_cell": false, "toc_section_display": "block", "toc_window_display": false } }, "nbformat": 4, "nbformat_minor": 4 }