{ "cells": [ { "cell_type": "markdown", "metadata": {}, "source": [ "### 一.模型结构\n", "线性回归算是回归任务中比较简单的一种模型,它的模型结构可以表示如下: \n", "\n", "$$\n", "f(x)=w^Tx^*\n", "$$ \n", "这里$x^*=[x^T,1]^T$,$x\\in R^n$,所以$w\\in R^{n+1}$,$w$即是模型需要学习的参数,下面造一些伪数据进行演示:" ] }, { "cell_type": "code", "execution_count": 1, "metadata": {}, "outputs": [], "source": [ "import numpy as np\n", "#造伪样本\n", "X=np.linspace(0,100,100)\n", "X=np.c_[X,np.ones(100)]\n", "w=np.asarray([3,2])\n", "Y=X.dot(w)\n", "X=X.astype('float')\n", "Y=Y.astype('float')\n", "X[:,0]+=np.random.normal(size=(X[:,0].shape))*3#添加噪声" ] }, { "cell_type": "code", "execution_count": 2, "metadata": {}, "outputs": [], "source": [ "Y=Y.reshape(100,1)" ] }, { "cell_type": "code", "execution_count": 3, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "Text(0,0.5,'Y')" ] }, "execution_count": 3, "metadata": {}, "output_type": "execute_result" }, { "data": { "image/png": "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\n", "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "import matplotlib.pyplot as plt\n", "%matplotlib inline\n", "plt.scatter(X[:,0],Y)\n", "plt.plot(np.arange(0,100).reshape((100,1)),Y,'r')\n", "plt.xlabel('X')\n", "plt.ylabel('Y')" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### 二.损失函数\n", "利用等式$y=3x+2$我造了一些伪数据,并给$x$添加了一些噪声数据,线性回归的目标即在只有$x,y$的情况下,求解出最优解:$w=[3,2]^T$;可以通过MSE(均方误差)来衡量$f(x)$与$y$的相近程度: \n", "\n", "$$\n", "L(w)=\\sum_{i=1}^m(y_i-f(x_i))^2=\\sum_{i=1}^m(y_i-w^Tx_i^*)^2=(Y-X^*w)^T(Y-X^*w)\n", "$$ \n", "\n", "这里$m$表示样本量,本例中$m=100$,$x_i,y_i$表示第$i$个样本,$X^*\\in R^{m \\times (n+1)},Y\\in R^{m\\times 1}$,损失函数$L(w)$本质上是关于$w$的函数,通过求解最小的$L(w)$即可得到$w$的最优解:\n", "\n", "$$\n", "w^*=arg \\min_{w}L(w)\n", "$$\n", "\n", "\n", "#### 方法一:直接求闭式解\n", "\n", "而对$\\min L(w)$的求解很明显是一个凸问题(海瑟矩阵${X^*}^TX^*$正定),我们可以直接通过求解$\\frac{dL}{dw}=0$得到$w^*$,梯度推导如下: \n", "\n", "$$\n", "\\frac{dL}{dw}=-2\\sum_{i=1}^m(y_i-w^Tx_i^*)x_i^*=-2{X^*}^T(Y-X^*w)\\\\\n", "$$ \n", "令$\\frac{dL}{dw}=0$,可得:$w^*=({X^*}^TX^*)^{-1}{X^*}^TY$,实际情景中数据不一定能满足${X^*}^TX$是满秩(比如$m" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "#可视化\n", "plt.scatter(X[:,0],Y)\n", "plt.plot(X[:,0],X.dot(w),'r')\n", "plt.xlabel('X')\n", "plt.ylabel('Y')" ] }, { "cell_type": "code", "execution_count": 6, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "Text(0,0.5,'loss')" ] }, "execution_count": 6, "metadata": {}, "output_type": "execute_result" }, { "data": { "image/png": "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\n", "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "#loss变化\n", "plt.plot(losses)\n", "plt.xlabel('iterations')\n", "plt.ylabel('loss')" ] }, { "cell_type": "code", "execution_count": 7, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "array([[2.97642542],\n", " [2.9148446 ]])" ] }, "execution_count": 7, "metadata": {}, "output_type": "execute_result" } ], "source": [ "#当然也可以直接求显式解\n", "w=np.linalg.pinv(X).dot(Y)\n", "w" ] }, { "cell_type": "code", "execution_count": 8, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "Text(0,0.5,'Y')" ] }, "execution_count": 8, "metadata": {}, "output_type": "execute_result" }, { "data": { "image/png": "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\n", "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "#可视化\n", "plt.scatter(X[:,0],Y)\n", "plt.plot(X[:,0],X.dot(w),'r')\n", "plt.xlabel('X')\n", "plt.ylabel('Y')" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### 四.问题讨论\n", "在上面的梯度下降的例子中存在一个问题,$w_1$基本能收敛到3附近,而$w_2$却基本在0附近,很难收敛到2,说明$w_1$比$w_2$更容易收敛($w=[w_1,w_2]^T$),这很容易理解,模型可以写作:$f(x)=x*w_1+1\\cdot w_2$,如果$x$量纲比1大很多,为了使$f(x)$变化,只需更新少量的$w_1$就能达到目的,而$w_2$的更新动力略显不足;可以考虑用如下方式: \n", "\n", "(1)对输入$X$进行归一化,使得$x$无量纲,$w_1,w_2$的更新动力一样(后面封装代码时添加上),如下图;\n", "![avatar](./source/01_归一化对梯度下降的影响.png)\n", "\n", "(2)梯度更新时,$w_1,w_2$使用了一样的学习率,可以让$w_1,w_2$使用不一样的学习率进行更新,比如对$w_2$使用更大的学习率进行更新(可以利用学习率自适应一类的梯度下降法,比如adam),如下图: \n", "![avatar](./source/01_adam.png)\n", "### 五.封装与测试\n", "接下来简单封装线性回归模型,并放到ml_models.linear_model模块便于后续使用;" ] }, { "cell_type": "code", "execution_count": 9, "metadata": {}, "outputs": [], "source": [ "class LinearRegression(object):\n", " def __init__(self, fit_intercept=True, solver='sgd', if_standard=True, epochs=10, eta=1e-2, batch_size=1):\n", " \"\"\"\n", " :param fit_intercept: 是否训练bias\n", " :param solver:\n", " :param if_standard:\n", " \"\"\"\n", " self.w = None\n", " self.fit_intercept = fit_intercept\n", " self.solver = solver\n", " self.if_standard = if_standard\n", " if if_standard:\n", " self.feature_mean = None\n", " self.feature_std = None\n", " self.epochs = epochs\n", " self.eta = eta\n", " self.batch_size = batch_size\n", "\n", " def init_params(self, n_features):\n", " \"\"\"\n", " 初始化参数\n", " :return:\n", " \"\"\"\n", " self.w = np.random.random(size=(n_features, 1))\n", "\n", " def _fit_closed_form_solution(self, x, y):\n", " \"\"\"\n", " 直接求闭式解\n", " :param x:\n", " :param y:\n", " :return:\n", " \"\"\"\n", " self.w = np.linalg.pinv(x).dot(y)\n", "\n", " def _fit_sgd(self, x, y):\n", " \"\"\"\n", " 随机梯度下降求解\n", " :param x:\n", " :param y:\n", " :param epochs:\n", " :param eta:\n", " :param batch_size:\n", " :return:\n", " \"\"\"\n", " x_y = np.c_[x, y]\n", " # 按batch_size更新w,b\n", " for _ in range(self.epochs):\n", " np.random.shuffle(x_y)\n", " for index in range(x_y.shape[0] // self.batch_size):\n", " batch_x_y = x_y[self.batch_size * index:self.batch_size * (index + 1)]\n", " batch_x = batch_x_y[:, :-1]\n", " batch_y = batch_x_y[:, -1:]\n", "\n", " dw = -2 * batch_x.T.dot(batch_y - batch_x.dot(self.w)) / self.batch_size\n", " self.w = self.w - self.eta * dw\n", "\n", " def fit(self, x, y):\n", " # 是否归一化feature\n", " if self.if_standard:\n", " self.feature_mean = np.mean(x, axis=0)\n", " self.feature_std = np.std(x, axis=0) + 1e-8\n", " x = (x - self.feature_mean) / self.feature_std\n", " # 是否训练bias\n", " if self.fit_intercept:\n", " x = np.c_[x, np.ones_like(y)]\n", " # 初始化参数\n", " self.init_params(x.shape[1])\n", " # 训练模型\n", " if self.solver == 'closed_form':\n", " self._fit_closed_form_solution(x, y)\n", " elif self.solver == 'sgd':\n", " self._fit_sgd(x, y)\n", "\n", " def get_params(self):\n", " \"\"\"\n", " 输出原始的系数\n", " :return: w,b\n", " \"\"\"\n", " if self.fit_intercept:\n", " w = self.w[:-1]\n", " b = self.w[-1]\n", " else:\n", " w = self.w\n", " b = 0\n", " if self.if_standard:\n", " w = w / self.feature_std.reshape(-1, 1)\n", " b = b - w.T.dot(self.feature_mean.reshape(-1, 1))\n", " return w.reshape(-1), b\n", "\n", " def predict(self, x):\n", " \"\"\"\n", " :param x:ndarray格式数据: m x n\n", " :return: m x 1\n", " \"\"\"\n", " if self.if_standard:\n", " x = (x - self.feature_mean) / self.feature_std\n", " if self.fit_intercept:\n", " x = np.c_[x, np.ones(shape=x.shape[0])]\n", " return x.dot(self.w)\n", "\n", " def plot_fit_boundary(self, x, y):\n", " \"\"\"\n", " 绘制拟合结果\n", " :param x:\n", " :param y:\n", " :return:\n", " \"\"\"\n", " plt.scatter(x[:, 0], y)\n", " plt.plot(x[:, 0], self.predict(x), 'r')" ] }, { "cell_type": "code", "execution_count": 10, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "w (array([2.97494802]), array([[3.14004069]]))\n" ] }, { "data": { "text/plain": [ "9.198087733141367" ] }, "execution_count": 10, "metadata": {}, "output_type": "execute_result" } ], "source": [ "#测试\n", "lr=LinearRegression(solver='sgd')\n", "lr.fit(X[:,:-1],Y)\n", "predict=lr.predict(X[:,:-1])\n", "#查看w\n", "print('w',lr.get_params())\n", "#查看标准差\n", "np.std(Y-predict)" ] }, { "cell_type": "code", "execution_count": 11, "metadata": {}, "outputs": [ { "data": { "image/png": "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\n", "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "#可视化结果\n", "lr.plot_fit_boundary(X[:,:-1],Y)" ] }, { "cell_type": "code", "execution_count": 12, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "w (array([2.97642542]), array([[2.9148446]]))\n" ] }, { "data": { "text/plain": [ "9.197986388759377" ] }, "execution_count": 12, "metadata": {}, "output_type": "execute_result" } ], "source": [ "#测试\n", "lr=LinearRegression(solver='closed_form')\n", "lr.fit(X[:,:-1],Y)\n", "predict=lr.predict(X[:,:-1])\n", "#查看w\n", "print('w',lr.get_params())\n", "#查看标准差\n", "np.std(Y-predict)" ] }, { "cell_type": "code", "execution_count": 13, "metadata": {}, "outputs": [ { "data": { "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXoAAAD8CAYAAAB5Pm/hAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMS4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvNQv5yAAAIABJREFUeJzt3Xt4VOW1x/HvIgQNWg0KVQhoUBHFWi6m4imtRWiLWoSIWi+tWm/UVltFRC61RetRQERqq3JE8QheUIoYMbVSC7RWjqhBBKFARaVKQKFy8UKAXN7zx+yZzJ7MLclMkpn8Ps/Dk+w9e/bs6diVNWu/73rNOYeIiGSvNs19ASIikl4K9CIiWU6BXkQkyynQi4hkOQV6EZEsp0AvIpLlFOhFRLKcAr2ISJZToBcRyXJtm/sCADp27OgKCwub+zJERDLKihUr/uOc65TouBYR6AsLCykrK2vuyxARyShm9u9kjlPpRkQkyynQi4hkOQV6EZEsp0AvIpLlFOhFRLKcAr2ISJZToBcRyXIK9CIizeHLL+HMM2HFirS/lAK9iEhTu+8+OPhgWLQIpk9P+8u1iJmxIiKtwtat0KVL7fa118KMGWl/WWX0IiJN4brr/EG+vLxJgjwo0IuIpNeaNWAGDz4Y2J42DZzzB/00U+lGRCQdamrgu9+FpUsD2zk5sGtXoDbfxJTRi4ik2ssvBwJ7MMg/+yxUVTVLkAdl9CIiqbN3Lxx9NGzbFtj++tcDwyfbNm+oVUYvIpIKs2ZBXl5tkF++HFatavYgD8roRUQa59NPoWPH2u2LLoKnngrcgG0hlNGLiDTUhAn+IP/++zB3bosK8qBALyJSfxs3BoL5pEmB7YkTA0Mmu3dv3uuKIWHpxswOBF4BDvCOn++cm2hm3YGngcOAt4BLnXP7zewAYA5wCvApcKFzblOarl9EJG1KVpYzddEGtuyqoEt+HmO+fzzFv70enn++9qAdO6BDh+a7yCQkk9HvAwY553oDfYAzzew0YAow3TnXA9gJXOUdfxWw0zl3HDDdO05EJKOUrCxn/IJ3KN9VgQOOXLOC4lO61Qb5OXMCWXwLD/KQREbvnHPAF95mrvfPAYOAS7z9s4HbgBnAcO93gPnA/WZm3nlERDLC1EUbqKispm11FX+Z9XOO2bkFgI8P/SpHbvsI2rVr1PnrfFsY0pPivgWpuPQ6khp1Y2Y5wArgOOAB4D1gl3OuyjtkMxC8wgLgIwDnXJWZ7QYOB/6TwusWEUmrLbsq+PrWf7Fwzk2hfRddfBevH/V1PkhBkB+/4B0qKqsBKN9VwfgF7wCkJdgndTPWOVftnOsDdAVOBU6Mdpj3M9rt5jrZvJmNNLMyMyvbvn17stcrIpJ++/bxmzefZv4TtwDw9+79KLzlBZYf9XUcMGDyEkpWljf49MFvC+EqKquZumhDY646pnqNo3fO7TKzvwGnAflm1tbL6rsCW7zDNgPdgM1m1hY4FNgR5VwzgZkARUVFKuuISEo0uiTy6qtwzTVcsX49C782iNvOuIod7Q/1HdLYDHzLrop67W+shBm9mXUys3zv9zzgu8A6YClwvnfY5UDwNvRCbxvv8SWqz4tIU4i8gRoMyEll37t3w89+Bt/+NlRUwEsvUTNnDnldjox6eGMy8C75efXa31jJlG46A0vNbDXwJvCyc64UGAvcZGYbCdTgZ3nHzwIO9/bfBIxL/WWLiNQVqyRy+wtrGTB5Cd3H/Sl62aWkBHr1gpkzYdSoQGvhIUMo7lvAsnGDotajoeEZ+JghPcnLzfHty8vNYcyQng06XyLJjLpZDfSNsv99AvX6yP17gQtScnUiIvUQK/Du3FPJzj2VQETZ5QiDX/wCFiwINCArKYFvfKPO87vk51Ee5dwNzcCD5Z4WNepGRCQTxArIkfbur2T9HffCklmBjpN33QU33wy5uVGPHzOkp2+UDDQ+Ay/uW5C2wB5JgV5EssYZJ3TiieUfxj3mmE83M2nR/fT/aA0MHAgPPQTHHx/3OU2dgaeaAr2IZJR4o2qWro89VLttdRUj31jADcvmsq9tOyaNGM34+VOTbkDWlBl4qinQi0jGSDTRKFaN/p4/Tef8NYsB+FPPAUw58+fcdPl3WlyXyXRRoBeRjBFvVM3URRvqzMzsV76OBU+MCW3PKhrOo+f9MqPKLqmgQC8iGSOZUTVBm6YM9R+0di1X9eoV6r7YmqgfvYhkjGSGM5760Zq6Qd65wDj5VkoZvYhkjGjDHIPM1fDB3cN8+77100f4tFMBk1aWt6pSTSRl9CKSMYr7FjBpxMkU5OdhQEF+Hvl5uVyw+mVfkJ8+4BIKx5ayOf9IKiqrufGZtxvdiCyTKaMXkRYr1lDKUHa+Zw8cdJDvOT1ufo7KnLoTn9LdCrglU0YvIi1SwgZlN93kC/K/vvjXFI4tjRrkg9LZCrglU0YvIs0qVtYeayjl//5xGcX9LvSfpKaGO8w4JWKcfTTpagXckinQi0iziTcBKlpAfubJsfTfvLZ2xxtv+JqQhbcqiNXzJl2tgFsyBXoRaRLRMvd4Ky2FNyj72scbKZ19Y+1B3/wmLFsW9XWCNfzIPyKQ3lbALZkCvYikXazMPVaJZcuuCqZf2Ifxz65m7Z0/oE3YnNeXXnqTM4cUJXzNTG9ElkrWEhZ/KioqcmVlZc19GSKSJgMmL4laSskxozpKDCrIz2PZcTvggtqlLf74X+eS+8AfAAXvIDNb4ZxL+FdPGb2IpF2sG6DVzpGXm+PL7A9pU8Oy8YP9B37xBRccdFDCpmYSnQK9iKRdrAVBCsJq9Vt2VTBhxXyu+etjocdX/vpurm9/Clvu+Btd8vPYs78qZk1fgT42BXoRSbt4KzQV9y2g+KgDoWNH33NKyj5kfMlaKrw/EPFWjmqNQybrQxOmRCRlSlaWR12EO1rrgkkjTg5k4SNG+IP80qXgHFNffjfuePhwrXHIZH0ooxeRlEhUP6+zQtP69f6FP449FjZuDG0mm6W31iGT9aGMXkRSIt6Y+HAlK8sDAf7EE2t3vvuuL8hD7Cw9Py83+jcDiSlhoDezbma21MzWmdlaM7vB23+bmZWb2dvev7PDnjPezDaa2QYzG5LONyAiLUOsDDy8tv76tEco7tc1tF3S6zuceOufKfm8blAfM6Qnebk5vn15uTncNuwklo0bxAeTf8CycYMU5JOQTOmmChjtnHvLzL4CrDCzl73Hpjvn7gk/2Mx6ARcBJwFdgL+a2fHOueSKbSKSkWKNrDECN1aLv3E0/cP29//5Y3zylY4QY9SMJjylTsJA75zbCmz1fv/czNYB8f6XHg487ZzbB3xgZhuBU4HXUnC9IpJmsZqMJTJmSE9GPfN2nXVbH5l/O4OnvBnavnPglTzcf4TvmFjfBurU9aVB6lWjN7NCoC/wurfrejNbbWaPmlkHb18B8FHY0zYT/w+DiLQQCVsDx1Hct8AX5A//chebpgxl8Hu1Qf70//5LnSAPGjWTbkkHejM7GHgWuNE59xkwAzgW6EMg458WPDTK0+vMcTazkWZWZmZl27dvr/eFi0jqJbqhGmv4ZFCBF7A3TRnKivt/HNo/a/Bl4Bw3nd0rat1do2bSK6nhlWaWSyDIP+mcWwDgnPsk7PGHgVJvczPQLezpXYEtked0zs0EZkKg101DLl5EUitWCWXLroqk2g9MPLk93x/qb19w4q1/ZtKIk33Hqe7etBI2NTMzA2YDO5xzN4bt7+zV7zGzUUB/59xFZnYS8BSBunwXYDHQI97NWDU1E2lewbp8rNmnwUw9VhuDZeMGQe/esHp1aP+V509kwymnK5CnUSqbmg0ALgXeMbO3vX0TgIvNrA+Bsswm4KcAzrm1ZjYP+CeBETvXacSNSMsVrW97uGBpZdQzb0d9vPOaFWBhWfyQIfDSSzyajouVBklm1M2rRK+7vxjnOXcCdzbiukSkiUSrywcVRCzt58vonWPT3ef4n/Dxx3DEEb5dDR3FI6mjmbEirVysuryBb0JS+ASmc9cs8Qf5CRPAuahBvqGjeCR11OtGpBUrWVlOmxiLf0QOeSzuW0DbLz5n6Okn+g+sqIADD4x6/nijeJTVNx1l9CKtVDDbjhbkow55NPMH+SeeCGTxMYI8xB/FI01HgV6klYpVm29jcEDbNox65m0GTF7C3+e84O8yCVBTAz/6UcLXiDURShOkmpYCvUgrFSurrnGwq6ISBywbP5jvXD4s9NiGq28IZPGRgT+GWI3JNEGqaSnQi7RS8bLqUf94gk1Thvr2FY4tpfjIM+t1IzXugiPSZBJOmGoKmjAlkhr1GcoYdfx8lCGTvzznZhb2GhjaDk2QkmaXyglTIpIBkmlREC6yHcHdr8zigtee8x1TOLa0zvN0IzXzKNCLZIlEQxljZfvFvTrWGTlz2s8e4+ND/It1B+lGauZRjV4kSyTTkCxy4tKnvYv8Qb64mJK3NpNzVKAvYeQtV91IzUyq0YtkiQGTl8RsOgb+hmRHfvYfls/4if/Aykpo6/+SH/4tIL99Ls7B7opKtTJoIZKt0SujF8kS8YYyhmf7m6YM9Qf5e+8NDJlsW7eSW9y3gGXjBjH9wj7srawJDbtUK4PMokAvkiXiDWXskp9H/w/fqTNksu/tixiwr3fMhUSCEi1IIi2bbsaKZJFYa6wuG+9fDOTii+6i7JjesLeKnXsqgfijdNTKILMpoxfJAjGX+LvvvjqzWLuPLeXD3v05qF1bKmv89+hiZelqZZDZlNGLZLho4+d/Nf9tivt19R/473/DUUfxgbfZfdyfop4vWpY+ZkjPOpOrNAIncyjQi2S4yPr5g8/dxdn/+r/Q9s5evemwtu7qUF3y86KO0omWpWut18ymQC+S4YIZePv9Ffxz+gW+x3qOXkCbvDwmrSyvE5TPOKETTy7/kPDiTbwsPVb9X1o+1ehFMlyX/Dy+9vFGX5Cf0f98CseWsq9tu6h195KV5Ty7otwX5A047xQF82ykjF4kk+3Zw2Pr/sgxj88EYG/bdpxw07N1bsBG1t1vf2FtneGSDli6fntaL1eahwK9SKb661/hpz+lx/vvs+ncS7j25B+yfl/0/0uH191LVpaHhlRG0nDJ7KTSjUim+fRTuOIK+N73ICcHli6lcMGTXFtcRH5ebp3DI+vu8SY5abhkdkoY6M2sm5ktNbN1ZrbWzG7w9h9mZi+b2bvezw7efjOz35vZRjNbbWb90v0mRFoF5+Dpp+HEEwPrtU6YAKtWwcCBoSGWuyr8mXqH9rl1FvqIl7VruGR2SiajrwJGO+dOBE4DrjOzXsA4YLFzrgew2NsGOAvo4f0bCcxI+VWLZIiYE5nq66OP4Jxz4OKLobAQysrgzjshL5CBx1r/tX27tnVursbK2vPzcnUjNkslDPTOua3Oube83z8H1gEFwHBgtnfYbKDY+304MMcFLAfyzaxzyq9cpIWL1Rq4XsG+uhr+8Afo1QuWLoXp0+G116B3b99h9WlREKv52W3DTkr+uiSj1KtGb2aFQF/gdeAI59xWCPwxAL7qHVYAfBT2tM3evshzjTSzMjMr275dd/ol+8RqBDZ63qrkgv3atfCtb8EvfwkDBgS2b7wxUJePUJ8WBVrHtfVJetSNmR0MPAvc6Jz7zGKvAh/tgTpN751zM4GZEOhHn+x1iGSKWFl2tXNxl/hj375AWWbyZDjkkEA9/pJL6gyZDFffFgWa/NS6JJXRm1kugSD/pHNugbf7k2BJxvu5zdu/GegW9vSuwJbUXK5I5og3giWY2dep3b/6KvTpA3fcARdeCOvWwY9+FDfIg7J0iS/hClMWSN1nAzucczeG7Z8KfOqcm2xm44DDnHO3mNkPgOuBs4H+wO+dc6fGew2tMCXZKLLZWDydqvcyb9PzdJ//OBx9NDz0EAwZ0gRXKZks2RWmkindDAAuBd4xs2BnpAnAZGCemV0FfAgE51+/SCDIbwT2AFfU89pFskIwmx49bxXVcRKqDfcUc0B1FdXWBkaNgt/+Fg4+uKkuU1qBhIHeOfcq0evuAIMjd7jAV4TrGnldIlkhGOyjZfa/WDaX0a8+Gdoe8eOpXHHphRQryEuKqQWCSAqFL6Yd2co3uP+AmirWTy32Pe9rN87jiwPaM3XRBtXVJeUU6EVSJNoCIOGja4r7FsDgwbBkSeg5rxT25bIL7whtR+sPL9JYCvQiKRJvAe3iI9tAly6+x467uYSqHP//BY3AHwxl9ZJKamomkiKxxs3/6rHf+IL8jCFXUzi2tE6Qh8CEk3hNx0QaQhm9SIpELs3XbdfH/OOhq33HnHjrnxMOt1SrYEk1BXqRMPFupiYSPjt18cPXcuyOzbUPrl3LgIUfU5FEEFerYEk1BXoRT6KbqYkU9y2g3c5POXtwbcOxrQOH0HnpSwBsmfNBwnPEa1sg0lAK9CKeuDdTwwJ91Ky/Txd44gnOHjWq9snbt9O5Y8fQZmRpJyjHjBrn6v0NQiRZuhkr4kmm1W+01sNv/uYeaNMGLrsMjj8e1qwJLBISFuQhdnvgaT/szQeTf8CycYMU5CUtlNGLeGJl3OE18/Cs/8DKvay/9/zaAw8+ONCUrE30/Cly4pQyeGkqCvQinmRa/Qb/ENy6+GGuLns+tP+6YWN54PnJCV9D7YGlOSjQi3iSybg7f7GD1x64zPe8wlteADNeHPcnZenSIinQi4SJm3Gffjqv/eMfoc1zLpvOO517hLbDlwsMnkukJdDNWJFEVqwILPzhBfmyghMpHFvqC/LhgiN1RFoKZfSS9Ro8Ccq5wPqsYb3kX3rpTX7xyjaojr9gj2a3SkuijF6yWrThkOMXvJN4ce5nnw2MngkG+RtuAOc4c0gRB7VLnB9pdqu0JMroJavFmwQVfNyX6ffuDLm5UFNT+4Qvv4T27UObuysq476mZrdKS6OMXrJarBJKMLMPz/QfeaiUT/v1rw3yDz8cyOjDgjzEz9a1KLe0RMroJavFazsQzPTbVVXys+V/5LrX5lFxQB7Mng2XXhq4ARtFrPH2CvDSUinQS1aLFZSD2w+UTOIHG5YBUNLrO9wx6BpWXPajuOfUDFfJNAr00mI1pmVwUKyg/MCMUl5++NrQcVed92sWH9efnBhZfLTzKrBLpkgY6M3sUWAosM059zVv323ANcB277AJzrkXvcfGA1cB1cAvnXOL0nDdkuUa2zI4XJ2g3LkzxR9/HNo8feTDfNihMwDVLv6wSZFMlExG/xhwPzAnYv9059w94TvMrBdwEXAS0AX4q5kd75yLv6SOSIRkWwbHEvXbwPa1MGRI6JgFJ53BTUNH+55XoGGRkoUSBnrn3CtmVpjk+YYDTzvn9gEfmNlG4FTgtQZfobRKybQMjiXy28DWHV9w6sB+8Nm20DGlf/8nv3r53xCngZlItmhMjf56M7sMKANGO+d2AgXA8rBjNnv7ROolmZbBQZHZ+579VaEgf/mKF7j9rw/VHjx1Ktx8M0OBqq8cohuq0io0NNDPAO4g0MfpDmAacCUQ7U5W1KKnmY0ERgIcddRRDbwMyVbJtAyG6LV8gMO/3MWK+3/sO/aYWxby/s3nhLZ1Q1VaiwYFeufcJ8HfzexhoNTb3Ax0Czu0K7AlxjlmAjMBioqKdAdMfJIdwhitlr9pylDf9pXn/YYlx51KjhklK8sV3KXVaVCgN7POzrmt3ua5wBrv94XAU2Z2L4GbsT2ANxp9ldIqJZNxh9fsv771Xyycc5Pv8cKxpaHfq51TC2FplZIZXjkXGAh0NLPNwERgoJn1IVCW2QT8FMA5t9bM5gH/BKqA6zTiRtIpWMuPzOKvOH8iS4/9Rp3j6zNyRyRbJDPq5uIou2fFOf5O4M7GXJS0bvWZKHVv7vv0n3KNb1/3saXRbwx51EJYWhvNjJUWJemJUjU1kJND/7DnDhw5k8rux+ISBHK1EJbWRt0rpUVJ1FYYgGuuCSwI4nn/sAIGTFrMjdeezbJxg+JOetJYeWmNlNFLixJ3otRnn8Ghh/r29xr1R/a0y4OwzD/a0EyADu1zmXjOSarPS6ujQC8tRsnKctqYRe03s/x/roAp20PbC/oP46aBI33HBDP/ZeMGAeouKRKkQC8tQrA2Hxnku+8oZ+nDP/UfXFPD6PEvRj1P8BuBJkOJ1FKglyYVa0RNtNr8E0//im/9e1Xtjscfhx8HZrvWp0WCSGunQC9NJt6ImvDa/CF7v2D1fRf5nxyR6SfbIkFENOpGmlC8ETVd8vPAOdZPG+EL8meNerxOkIdAaWbSiJMpyM/D0FqtIvEoo5cmE29EzdwjPuG0u68K7bvsgtt55ZhTyG0Tuz+N6vAiyVGglyaT3z6XnXsqffvaVlex8Z5i377QkEmgssapZYFIIynQS8rEa11QsrKcL/ZW+Y6f+NeHuGLFC6HtW7//c57oe3ad86plgUjjKNBLSiRqXTB10QYqawK19iM/+w/LZ/zEf4Lqapbe/TfQSBqRlNPNWEmJRK0LgkMh//D8FF+Q//X3fha42dqmDWOG9CQvN8d3Do2kEWk8ZfSSEvFutJasLOeYHeUsiZj4VDi2lIL8PO7wtpNdbERE6keBXlIi3gSm/oNOYcmu0KJkDLr6f3j/8K4AnHFCJ9/xGkkjknoq3UhKRCu7DNz8DsvGD6azF+RLT/g2hWNLQ0Ee4NkV5ZSsLG/SaxVpbZTRS0qEl1227vyS9+8e5nu89y/nsjvvK3WeV1FZzeh5q3znEJHUUkYvKVPct4BlHd71B/m77qLkrc3sPyQ/5vOCa7kqsxdJD2X0Um9Rx8uffATcey+MHVt74P79kJtLcDrU6HmrorYgBq3lKpJOyuilXoLj5ct3VeAIDJt8/IHn2HVy30CQLyiAF18MDJnMzQ09r7hvAdN+2LtOHT+cJkaJpIcyeqnXYtzh4+Xz9u9l9D8e5ycrXmDXwfkwfz6MGAFmUZ8bPGeszF4To0TSQ4G+lUt6MW5PMOsumTOKPlvfBeCp3kOYMvAKVp13XsLXC55TLYZFmk7C0o2ZPWpm28xsTdi+w8zsZTN71/vZwdtvZvZ7M9toZqvNrF86L14aL6nFuMN897NNbJoyNBTkn/3aICac+QvaHNYh6ddUi2GRppVMRv8YcD8wJ2zfOGCxc26ymY3ztscCZwE9vH/9gRneT2mhYtXFy70ZraHg67UpeDjsmG9cN4ftBx8GwBd7q2K2E45GE6NEmk7CjN459wqwI2L3cGC29/tsCA2sGA7McQHLgXwz65yqi5XUi1cXDw15nDsX2tT+p/LYNy+gcGxpKMhDbTthEWl5Gjrq5gjn3FYA7+dXvf0FwEdhx2329kkLFW1Ga1B1xV6K+3WFSy6p3blnD7d/+/Kox2vUjEjLlOrhldGGW0QdOG1mI82szMzKtm/fnuLLkGQF6+WRNk0Zyr+mnRvaHn/OTZS8tRny8mJ+C9CoGZGWqaGB/pNgScb7uc3bvxnoFnZcV2BLtBM452Y654qcc0WdOnWKdoikSMnKcgZMXkL3cX9iwOQldWagFvctoMAL0id98h6bpgz1PV54ywvM7TUoVJpRO2GRzNLQ4ZULgcuByd7P58P2X29mTxO4Cbs7WOKR5pHs8MkxQ3oGyjRhHjp1BJPOuDK0HSzNqJ2wSGZJGOjNbC4wEOhoZpuBiQQC/Dwzuwr4ELjAO/xF4GxgI7AHuCIN1yz1EG/4ZCgwr1lDcT9/+aZwbGmdc4WXZjRqRiRzJAz0zrmLYzw0OMqxDriusRclqRNvQRAA8vNh9+7aB5Yv59at7WH5h3WeE9k7XkQyg3rdZLlYN0gHv/dmoFVBMMhfeWVgrHz//ixdH/3meKz9ItKyqQVClhszpKevRp9TU817U4f7jikaM59bLz4tNBki4bcAEckoyuizXHi7gavfeM4X5G8ffA2FY0v5T5sDfZOdNHxSJLsoo28Fio85mOLx/lsqx4x5npo2tUMkw9d7jfwWABo+KZLJlNFnu8svD9xw9VxxwW0Uji31BXmAnLDWwmo6JpJdlNFniPr0jAfg/ffh2GNrt484gpI/r2DpM29HPTyyP7yGT4pkD2X0GSDaqk5x11jt2dMf5NeuhY8/jtt0rED1d5GspYw+A8Sa9HT7C2t9Wf7kjjv59jXn1x40fDiUlIQ2442aUf1dJHsp0GeAWAF6555Kdu6pxFwNyyJutrJ9O3Ts6NvVJT/Pd9M1KD8vV2UakSym0k0GiDes8Yer/sIHdw8LbT/y3csDE5+8IB/e0OzLfVXk5vgbjObl5nDbsJPSc+Ei0iIoo88A0YY7Hli5l/X3nu87rsfNz1GVk8vV3nZkQ7NdFZXktjE6tM9l155KNSMTaSUU6FuARCNqIrtFTn71f7lw2bOhx38+fBwvnvAtwH9TNVptv7LG0b5dW1b+5vvpfEsi0oIo0DezZNsIF/ctoLiTg27dfM8vvOWFQM8a6k5qUisDEQHV6JtdvDbCPma+IH/VdQ9y63OrKejQPuakJrUyEBFQRt/sEmbd8+bBhReG9q/s3JNzL5sGQN6K8rgzVtXKQERAgb7JxKrDxxry2OXQA0MlmaAhV97Phk6Foe06C4hE0EpQIgIK9E0iXh0+WtY98W+PcsXrC2pPcOihdL/2yairrCeqt6uVgYgo0KdYtMw9Xh1+2bhBQCDr3v3Jp6z53Q/9J9yxAzp0oMvkJdEzf9XbRSQBBfoUipW5Rwb5oPDFtosv+A68917tgxddBHPnhs775b6qOs9XvV1EkqFAn0KxMvccszrdIcHLxjdtgu7d/Q9UVUFOoI1w5B+PoA7tc5l4zkkqy4hIQgr0KRSrXl7tHHm5OXVGvywbPxjGhx344IOUnDaMqVP/Hir97NlfFfUbQft2bRXkRSQpGkefQrHq5cEx7sGFPIb/55+s+++z/Ac5R8lpw+q0I965pzLqOTXpSUSS1aiM3sw2AZ8D1UCVc67IzA4DngEKgU3AD51zOxt3mS1HvHYF8cath0a/RAyZ5NVXYcAAIHrpJxbdhBWRZKUioz/DOdfHOVfkbY8DFjvnegCLve2skGgBkLhL8N19d90g71woyEPyWbpuwopIfaSjRj8cGOj9Phv4GzA2Da+TNrGy9njDJINZfZ1x65VAlGRVAAAJvUlEQVSVdQN8eTl06VLndeP1iz/ogLaa9CQiDdLYQO+Av5iZAx5yzs0EjnDObQVwzm01s69Ge6KZjQRGAhx11FGNvIzUiTe5qd5Nws4+G/7859rt00+Hv/895mvHKv3cNkyja0Sk4Rob6Ac457Z4wfxlM1uf7BO9PwozAYqKiqJN+mwW8bL2mO0KIuvln30Ghx7q37dvH7RrF/N1g98iwodjFih7F5EUaFSN3jm3xfu5DXgOOBX4xMw6A3g/tzX2IptSvKx9zJCe5OXm+PbXqZdffbU/yE+cGKjFJwjywdo/1A7HVJAXkVRocKA3s4PM7CvB34HvA2uAhcDl3mGXA8839iKbUrzWvnFvtr7xRqAWP2tW6Dndx5Yy4MDTQzdrY0m6VbGISAM0pnRzBPCcBW40tgWecs69ZGZvAvPM7CrgQ+CCxl9m00nU2jdqk7CIm63Dr3mQVYcF7jvEWkgknBYIEZF0anCgd869D/SOsv9TYHBjLqo5JWrtGz4i55qNf2PCs/f4nj9g0uI6dfxE7YSTrv2LiDSAWiBEEau1b7CWvnd/JR/cPcz32Mslr3DbusqoARviZ+daIERE0kmBvh6mLtrAxIXTuWj1X0L7/p1/JMWjZrO37Mu4s1rjZedaIERE0qlVBvp4bQxi+uyzQBOyMCfdOI8vD2gPMfrRBCWTnWuBEBFJl1bX1CxRG4Oounb1DZl8vO/ZFI4tDQT5BKIt2i0i0pRaXUafTBuDkPXr4cQTfbt6jC+lsqZ2OzfHOKhdW3ZV1M3qC/LzQitIiYg0l1aX0Sc9lNHMH+TnzKHkrc1AZGMyGNq7c+KJVCIizaTVBfr89rlR94dulr79dvQuk5deytRFG6is8XdrqKxxLF2/PfZEKhGRZtaqSjclK8v5Ym/dtVdzc4wx3z8e8vJg797aBz78ELp1C23G+zagm6ki0lK1qow+WkYOcM76f1B8SrfaIP+znwWy+LAgD/HbI4iItFStKqOPzMgPqNrPhmkj/Ad9/jkcfHDU52tik4hkolaV0Ydn3gseH+0L8k/0OYsBkxZT8u7umM+P29RMRKSFalUZ/ZghPRnz1JvMmv9b+m2p7Qx57JjnqW6TA0k0IFMtXkQyTavK6Iv3fsifZ9/I6ZtWAnDHoKspHFsaCPIetQcWkWyTVRl9zNYGW7fC6NHw9NN0OuyrXHneb1hy3Kkxz6P2wCKSTbIm0Mda6/XrkyZwzB/nBA66/nrO7ziEdysszpk0ikZEskvGBvrI7H3P/irfaJie2zex6NHra58wbRrcdBMbx/0p7nk1ikZEsk1GBvpo2XuQuRrmzp3AaR+tAaCyTQ65u3eFhkzGWuQD0GLcIpKVMvJmbLTGZADf+mAlH9w9LBTkf3ruBE64ZaFvyGSsBb5/d2Eflo0bpCAvIlknIzP6aBOf/u/Bn3B4xWcArOtUyNCf3BcYTeOcb8ikFvkQkdYmIwN9ePnlwlWLmPLSH0KPnXvZvazsfLzv+IrKakbPWwXUBnsFdhFpLTKydBMsvxTuKA8F+dKTBlKy4iPejgjyQdVeZh93gRERkSyUtkBvZmea2QYz22hm41J57mArAnfU0Uz79o85f8wTVD3+BMX9usYdGllRWc3tL6xN5aWIiLR45lzdbo6NPqlZDvAv4HvAZuBN4GLn3D+jHV9UVOTKyspS8tqRI3Ki+d2FfVS6EZGMZ2YrnHNFiY5LV0Z/KrDROfe+c24/8DQwPE2v5RPM9nMiFw8JoxYHItKapCvQFwAfhW1v9vY1ieK+BUz7Ye+Yj6vFgYi0JukK9NHSaV+NyMxGmlmZmZVt37495RdQ3LeA/LwEywaKiLQC6Qr0m4Hw5Zm6AlvCD3DOzXTOFTnnijp16pSWi7ht2ElatFtEWr10jaN/E+hhZt2BcuAi4JI0vVZMmhwlIpKmQO+cqzKz64FFQA7wqHOuWcY1anKUiLR2aZsZ65x7EXgxXecXEZHkZOTMWBERSZ4CvYhIllOgFxHJcgr0IiJZLi29bup9EWbbgX8neXhH4D9pvJyWojW8T73H7KD32HyOds4lnIjUIgJ9fZhZWTJNfDJda3ifeo/ZQe+x5VPpRkQkyynQi4hkuUwM9DOb+wKaSGt4n3qP2UHvsYXLuBq9iIjUTyZm9CIiUg8ZFejTuQ5tczGzbma21MzWmdlaM7vB23+Ymb1sZu96Pzs097U2lpnlmNlKMyv1trub2evee3zGzNo19zU2hpnlm9l8M1vvfZ7/lW2fo5mN8v47XWNmc83swGz4HM3sUTPbZmZrwvZF/ews4PdeHFptZv2a78qTkzGB3luH9gHgLKAXcLGZ9Wreq0qJKmC0c+5E4DTgOu99jQMWO+d6AIu97Ux3A7AubHsKMN17jzuBq5rlqlLnPuAl59wJQG8C7zVrPkczKwB+CRQ5575GoDPtRWTH5/gYcGbEvlif3VlAD+/fSGBGE11jg2VMoKcZ16FNJ+fcVufcW97vnxMIDgUE3tts77DZQHHzXGFqmFlX4AfAI962AYOA+d4hGf0ezewQ4HRgFoBzbr9zbhdZ9jkS6HibZ2ZtgfbAVrLgc3TOvQLsiNgd67MbDsxxAcuBfDPr3DRX2jCZFOibdR3apmBmhUBf4HXgCOfcVgj8MQC+2nxXlhK/A24Barztw4FdzrkqbzvTP89jgO3A/3rlqUfM7CCy6HN0zpUD9wAfEgjwu4EVZNfnGC7WZ5dxsSiTAn3CdWgzmZkdDDwL3Oic+6y5ryeVzGwosM05tyJ8d5RDM/nzbAv0A2Y45/oCX5LBZZpovBr1cKA70AU4iEAZI1Imf47JyLj/djMp0CdchzZTmVkugSD/pHNugbf7k+DXQe/ntua6vhQYAAwzs00ESm6DCGT4+V4JADL/89wMbHbOve5tzycQ+LPpc/wu8IFzbrtzrhJYAHyT7Pocw8X67DIuFmVSoA+tQ+vd1b8IWNjM19RoXq16FrDOOXdv2EMLgcu93y8Hnm/qa0sV59x451xX51whgc9tiXPuR8BS4HzvsEx/jx8DH5lZcOX5wcA/yaLPkUDJ5jQza+/9dxt8j1nzOUaI9dktBC7zRt+cBuwOlnhaLOdcxvwDzgb+BbwH/Kq5rydF7+lbBL72rQbe9v6dTaCGvRh41/t5WHNfa4re70Cg1Pv9GOANYCPwR+CA5r6+Rr63PkCZ91mWAB2y7XMEbgfWA2uAx4EDsuFzBOYSuO9QSSBjvyrWZ0egdPOAF4feITAKqdnfQ7x/mhkrIpLlMql0IyIiDaBALyKS5RToRUSynAK9iEiWU6AXEclyCvQiIllOgV5EJMsp0IuIZLn/B2KwxAohCjP2AAAAAElFTkSuQmCC\n", "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "#可视化结果\n", "lr.plot_fit_boundary(X[:,:-1],Y)" ] }, { "cell_type": "code", "execution_count": 14, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "w: [[2.97642542]] b: [2.9148446]\n" ] }, { "data": { "text/plain": [ "9.197986388759379" ] }, "execution_count": 14, "metadata": {}, "output_type": "execute_result" } ], "source": [ "#与sklearn对比\n", "from sklearn.linear_model import LinearRegression\n", "lr=LinearRegression()\n", "lr.fit(X[:,:-1],Y)\n", "predict=lr.predict(X[:,:-1])\n", "#查看w,b\n", "print('w:',lr.coef_,'b:',lr.intercept_)\n", "#查看标准差\n", "np.std(Y-predict)" ] }, { "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": "3.6.4" } }, "nbformat": 4, "nbformat_minor": 2 }