{ "cells": [ { "cell_type": "code", "execution_count": 1, "metadata": {}, "outputs": [], "source": [ "import numpy as np\n", "from scipy.optimize import minimize\n", "from sklearn.datasets import load_iris\n", "from sklearn.svm import LinearSVC as skLinearSVC" ] }, { "cell_type": "code", "execution_count": 2, "metadata": {}, "outputs": [], "source": [ "class LinearSVC():\n", " def __init__(self, C=1.0):\n", " self.C = C\n", "\n", " def _encode(self, y):\n", " classes = np.unique(y)\n", " y_train = np.full((y.shape[0], len(classes)), -1)\n", " for i, c in enumerate(classes):\n", " y_train[y == c, i] = 1\n", " if len(classes) == 2:\n", " y_train = y_train[:, 1].reshape(-1, 1)\n", " return classes, y_train\n", "\n", " @staticmethod\n", " def _cost_grad(w, X, y, C):\n", " X_train = np.c_[X, np.ones(X.shape[0])]\n", " z = np.dot(X_train, w)\n", " yz = y * z\n", " mask = yz <= 1\n", " cost = C * np.sum(np.square(1 - yz[mask])) + 0.5 * np.dot(w, w)\n", " grad = w + 2 * C * np.dot(X_train[mask].T, z[mask] - y[mask])\n", " return cost, grad\n", " \n", " def _solve_lbfgs(self, X, y):\n", " result = np.zeros((y.shape[1], X.shape[1] + 1))\n", " for i in range(y.shape[1]):\n", " cur_y = y[:, i]\n", " w0 = np.zeros(X.shape[1] + 1)\n", " res = minimize(fun=self._cost_grad, jac=True, x0=w0,\n", " args=(X, cur_y, self.C), method='L-BFGS-B')\n", " result[i] = res.x\n", " return result[:, :-1], result[:, -1]\n", "\n", " def fit(self, X, y):\n", " self.classes_, y_train = self._encode(y)\n", " self.coef_, self.intercept_ = self._solve_lbfgs(X, y_train)\n", " return self\n", "\n", " def decision_function(self, X):\n", " scores = np.dot(X, self.coef_.T) + self.intercept_\n", " if scores.shape[1] == 1:\n", " return scores.ravel()\n", " else:\n", " return scores\n", "\n", " def predict(self, X):\n", " scores = self.decision_function(X)\n", " if len(scores.shape) == 1:\n", " indices = (scores > 0).astype(int)\n", " else:\n", " indices = np.argmax(scores, axis=1)\n", " return self.classes_[indices]" ] }, { "cell_type": "code", "execution_count": 3, "metadata": {}, "outputs": [], "source": [ "X, y = load_iris(return_X_y=True)\n", "X, y = X[y != 2], y[y != 2]\n", "clf1 = LinearSVC().fit(X, y)\n", "clf2 = skLinearSVC(dual=False).fit(X, y)\n", "assert np.allclose(clf1.coef_, clf2.coef_, atol=1e-2)\n", "assert np.allclose(clf1.intercept_, clf2.intercept_, atol=1e-3)\n", "prob1 = clf1.decision_function(X)\n", "prob2 = clf2.decision_function(X)\n", "assert np.allclose(prob1, prob2, atol=1e-2)\n", "pred1 = clf1.predict(X)\n", "pred2 = clf2.predict(X)\n", "assert np.array_equal(pred1, pred2)" ] }, { "cell_type": "code", "execution_count": 4, "metadata": {}, "outputs": [], "source": [ "X, y = load_iris(return_X_y=True)\n", "clf1 = LinearSVC().fit(X, y)\n", "clf2 = skLinearSVC(dual=False).fit(X, y)\n", "assert np.allclose(clf1.coef_, clf2.coef_, atol=1e-1)\n", "assert np.allclose(clf1.intercept_, clf2.intercept_, atol=1e-2)\n", "prob1 = clf1.decision_function(X)\n", "prob2 = clf2.decision_function(X)\n", "assert np.allclose(prob1, prob2, atol=1e-1)\n", "pred1 = clf1.predict(X)\n", "pred2 = clf2.predict(X)\n", "assert np.array_equal(pred1, pred2)" ] } ], "metadata": { "kernelspec": { "display_name": "dev", "language": "python", "name": "dev" }, "language_info": { "codemirror_mode": { "name": "ipython", "version": 3 }, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.7.3" } }, "nbformat": 4, "nbformat_minor": 2 }