logistic_regression_from_scratch.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "from sklearn.datasets import load_digits\n",
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "np.random.seed(13)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Данные"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = load_digits()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "X = data['data']\n",
    "y = (data['target'] >= 5).astype(np.int32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Figure size 432x288 with 0 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 288x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "6\n"
     ]
    }
   ],
   "source": [
    "plt.gray()\n",
    "index = np.random.randint(0, len(X) - 1)\n",
    "plt.matshow(X[index].reshape(8, 8))\n",
    "plt.show()\n",
    "print(data.target[index])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train, X_test, y_train, y_test = train_test_split(X,\n",
    "                                                    y,\n",
    "                                                    train_size=0.77,\n",
    "                                                    shuffle=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((1383, 64), (414, 64))"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train.shape, X_test.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Логистическая регрессия"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Вычисление предсказания логистической регрессией для объекта $x_i \\in \\mathbb{R}^n$ :\n",
    "\n",
    "* $z = w^T x_i + b, \\quad w \\in \\mathbb{R}^n, b \\in \\mathbb{R}$\n",
    "* $a = \\sigma (z)$"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Сигмоида\n",
    "\n",
    "$$\\sigma(x) = \\frac{1}{1 + e^{-x}}$$\n",
    "\n",
    "![alt text](https://upload.wikimedia.org/wikipedia/commons/a/ac/Logistic-curve.png)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Логистическая фунция потерь:\n",
    "\n",
    "$$L_i(a) = -y_i \\cdot log(a) - (1 - y_i) \\cdot log(1 - a)$$"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Производные функции по параметрам:\n",
    "\n",
    "$$\\frac{\\partial L_i}{\\partial w} = \\frac{\\partial L_i}{\\partial a} \\cdot \\frac{\\partial a}{\\partial z} \n",
    "\\cdot \\frac{\\partial z}{\\partial w} = \\frac {a - y_i} {a \\cdot (1-a)} \\cdot a \\cdot (1-a) \\cdot x_i = (a-y_i) \\cdot x_i $$\n",
    "\n",
    "$$\\frac{\\partial L_i}{\\partial b} = \\frac{\\partial L_i}{\\partial a} \\cdot \\frac{\\partial a}{\\partial z} \n",
    "\\cdot \\frac{\\partial z}{\\partial b} = \\frac {a - y_i} {a \\cdot (1-a)} \\cdot a \\cdot (1-a) \\cdot 1 = a-y_i $$\n",
    "\n",
    "Используются для минимизации функции потерь $L$ с помощью градиентного спуска."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "def log_loss(y_true, y_pred):\n",
    "    return -np.sum(y_true * np.log(y_pred) + (1 - y_true) * np.log(1 - y_pred), axis=0) / len(y_true)\n",
    "\n",
    "def sigmoid(x):\n",
    "    return 1 / (1 + np.exp(-x))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Число объектов в обучающей выборке: 1383\n",
      "Размерность объекта: 64\n"
     ]
    }
   ],
   "source": [
    "m, n = X_train.shape\n",
    "print(f'Число объектов в обучающей выборке: {m}\\nРазмерность объекта: {n}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "class LogisticRegression:\n",
    "    \n",
    "    def __init__(self):\n",
    "        self.w = np.random.randn(n, 1) * 0.001\n",
    "        self.b = np.random.randn() * 0.001\n",
    "        self.report_every = 40\n",
    "\n",
    "    def train(self, X, y, learning_rate=0.005, epochs=40):\n",
    "        self.losses_train = []\n",
    "        self.losses_test = []\n",
    "        \n",
    "        for epoch in range(epochs):            \n",
    "            dw = np.zeros((n, 1))\n",
    "            db = 0\n",
    "            \n",
    "            for i in range(len(X)):\n",
    "                # forward propagation\n",
    "                z = X[i].reshape(1, n).dot(self.w) + self.b\n",
    "                a = sigmoid(z)[0][0]\n",
    "                \n",
    "                # gradient calculation\n",
    "                dw += (a - y[i]) * X[i].reshape(n, 1)\n",
    "                db += (a - y[i])\n",
    "                \n",
    "            dw /= len(X)\n",
    "            db /= len(X)\n",
    "            \n",
    "            # gradient step\n",
    "            self.w = self.w - learning_rate * dw\n",
    "            self.b = self.b - learning_rate * db\n",
    "            \n",
    "            # save loss for plotting\n",
    "            if epoch % self.report_every == 0:\n",
    "                self.losses_train.append(log_loss(y, self.predict(X)))\n",
    "                self.losses_test.append(log_loss(y_test, self.predict(X_test)))\n",
    "        \n",
    "\n",
    "    def train_vec(self, X, y, learning_rate=0.005, epochs=40):\n",
    "        self.losses_train = []\n",
    "        self.losses_test = []\n",
    "        \n",
    "        for epoch in range(epochs):            \n",
    "            Z = X.reshape(m, n).dot(self.w) + self.b\n",
    "            A = sigmoid(Z)\n",
    "            \n",
    "            dw = np.sum(X.reshape(m, n) * (A.reshape(m, 1) - y.reshape(m, 1)), axis=0) / len(X)\n",
    "            db = np.sum((A.reshape(m, 1) - y.reshape(m, 1)), axis=0) / len(X)\n",
    "            \n",
    "            # gradient step\n",
    "            self.w = self.w - learning_rate * dw.reshape(n, 1)\n",
    "            self.b = self.b - learning_rate * db\n",
    "            \n",
    "            # save loss for plotting\n",
    "            if epoch % self.report_every == 0:\n",
    "                self.losses_train.append(log_loss(y, self.predict(X)))\n",
    "                self.losses_test.append(log_loss(y_test, self.predict(X_test)))\n",
    "    \n",
    "\n",
    "    def predict(self, X):        \n",
    "        return np.array([sigmoid(x.reshape(1, n).dot(self.w) + self.b)[0][0] \n",
    "                         for x in X])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Используя цикл"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 7.37 s, sys: 152 ms, total: 7.52 s\n",
      "Wall time: 7.38 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "logreg = LogisticRegression()\n",
    "logreg.train(X_train, y_train, epochs=500)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    " #### Обучалось примерно 7 секунд"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "domain = np.arange(0, len(logreg.losses_train)) * logreg.report_every\n",
    "plt.plot(domain, logreg.losses_train, label='Train')\n",
    "plt.plot(domain, logreg.losses_test, label='Test')\n",
    "plt.xlabel('Epoch number')\n",
    "plt.ylabel('LogLoss')\n",
    "plt.legend();"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Точность на тестовой выборке: 89.13%\n"
     ]
    }
   ],
   "source": [
    "test_prediction = np.array(logreg.predict(X_test))\n",
    "test_accuracy = np.sum((test_prediction > 0.5) == y_test) / len(test_prediction)\n",
    "print(f'Точность на тестовой выборке: {round(test_accuracy * 100, 2)}%')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Точность на тестовой выборке: 90.96%\n"
     ]
    }
   ],
   "source": [
    "train_prediction = np.array(logreg.predict(X_train))\n",
    "train_accuracy = np.sum((train_prediction > 0.5) == y_train) / len(train_prediction)\n",
    "print(f'Точность на тестовой выборке: {round(train_accuracy * 100, 2)}%')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Используя векторизацию"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 1.25 s, sys: 219 ms, total: 1.46 s\n",
      "Wall time: 247 ms\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "logreg = LogisticRegression()\n",
    "logreg.train_vec(X_train, y_train, epochs=500)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "####  Обучалось примерно 0.26 секунд"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "domain = np.arange(0, len(logreg.losses_train)) * logreg.report_every\n",
    "plt.plot(domain, logreg.losses_train, label='Train')\n",
    "plt.plot(domain, logreg.losses_test, label='Test')\n",
    "plt.xlabel('Epoch number')\n",
    "plt.ylabel('LogLoss')\n",
    "plt.legend();"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Точность на тестовой выборке: 89.13%\n"
     ]
    }
   ],
   "source": [
    "test_prediction = np.array(logreg.predict(X_test))\n",
    "test_accuracy = np.sum((test_prediction > 0.5) == y_test) / len(test_prediction)\n",
    "print(f'Точность на тестовой выборке: {round(test_accuracy * 100, 2)}%')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Точность на тестовой выборке: 90.89%\n"
     ]
    }
   ],
   "source": [
    "train_prediction = np.array(logreg.predict(X_train))\n",
    "train_accuracy = np.sum((train_prediction > 0.5) == y_train) / len(train_prediction)\n",
    "print(f'Точность на тестовой выборке: {round(train_accuracy * 100, 2)}%')"
   ]
  }
 ],
 "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": "3.7.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}