simple.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import math\n",
    "import shutil\n",
    "import numpy as np\n",
    "from PIL import Image\n",
    "import matplotlib.pyplot as plt\n",
    "from utils.draw import draw_single"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "def generate(count):\n",
    "    X = np.random.randint(0, high=255, size=(count, 9))\n",
    "    Y = X.dot(np.array([1, 1, 1, 0, 0, 0, -1, -1, -1]))\n",
    "    Y[Y > 0] = 1\n",
    "    Y[Y < 0] = -1\n",
    "    return X, Y"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1296x504 with 120 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "draw_single(*generate(120))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "def train(n=512, epochs=31000, lr = 1e-3):\n",
    "    # weight vector\n",
    "    W = np.random.randn(9, 1)\n",
    "\n",
    "    # training loop\n",
    "    for t in range(epochs):\n",
    "        # get new training data\n",
    "        X, y = generate(n)\n",
    "        X = X / 255\n",
    "        y = y.reshape(n, 1)\n",
    "\n",
    "        # model function\n",
    "        h = X.dot(W)\n",
    "\n",
    "        # compute loss\n",
    "        loss = np.square(h - y).mean()\n",
    "\n",
    "        # compute accuracy\n",
    "        acc = (np.sign(h) == y).mean()\n",
    "\n",
    "        if t % 5000 == 0:\n",
    "            print('l: {:>8f}, a {:>.4f} (e {})'.format(loss, acc, t))\n",
    "\n",
    "        # grad + update\n",
    "        grad_w = 2 * X.T.dot(h - y) / n\n",
    "        W -= lr * grad_w\n",
    "\n",
    "    return W"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "l: 2.826098, a 0.5215 (e 0)\n",
      "l: 0.565322, a 0.8223 (e 5000)\n",
      "l: 0.412433, a 0.9297 (e 10000)\n",
      "l: 0.365760, a 0.9746 (e 15000)\n",
      "l: 0.347903, a 0.9922 (e 20000)\n",
      "l: 0.338252, a 1.0000 (e 25000)\n",
      "l: 0.373862, a 1.0000 (e 30000)\n",
      "\n",
      "Final W = \n",
      "\n",
      "[[ 1.13179384]\n",
      " [ 1.13089455]\n",
      " [ 1.13347194]\n",
      " [-0.00384232]\n",
      " [-0.0078364 ]\n",
      " [ 0.00724967]\n",
      " [-1.13251403]\n",
      " [-1.13009286]\n",
      " [-1.12999758]]\n"
     ]
    }
   ],
   "source": [
    "W = train()\n",
    "print('\\nFinal W = \\n\\n{}'.format(W))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "file_extension": ".py",
  "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": "3.6.10"
  },
  "mimetype": "text/x-python",
  "name": "python",
  "npconvert_exporter": "python",
  "pygments_lexer": "ipython3",
  "version": 3
 },
 "nbformat": 4,
 "nbformat_minor": 2
}