triple.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",
    "from utils.draw import draw_xy\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "def generate(count):\n",
    "    X = np.random.randint(0, high=255, size=(count, 9))\n",
    "    a = np.array([[1, 1, 1, 0, 0, 0, 0, 0, 0], \n",
    "                  [0, 0, 0, 1, 1, 1, 0, 0, 0], \n",
    "                  [0, 0, 0, 0, 0, 0, 1, 1, 1]])\n",
    "\n",
    "    Y = np.eye(3)[np.argmax(X.dot(a.T), axis=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_xy(*generate(120))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "def train(n=512, epochs=56000, lr = 1e-3):\n",
    "    # weight vector\n",
    "    W = np.random.randn(9, 3)\n",
    "    b = np.random.randn(1, 3)\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",
    "\n",
    "        # model function\n",
    "        h = X.dot(W) + b\n",
    "\n",
    "        # compute loss\n",
    "        loss = np.square(h - y).mean()\n",
    "\n",
    "        # compute accuracy\n",
    "        acc = (np.argmax(h, axis=1) == np.argmax(y, axis=1)).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",
    "        grad_b = 2 * np.sum(h - y, axis=0) / n\n",
    "        b -= lr * grad_b\n",
    "\n",
    "    return W, b"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "l: 2.856105, a 0.2734 (e 0)\n",
      "l: 0.203259, a 0.5996 (e 5000)\n",
      "l: 0.125143, a 0.8164 (e 10000)\n",
      "l: 0.109183, a 0.8926 (e 15000)\n",
      "l: 0.104883, a 0.9004 (e 20000)\n",
      "l: 0.099931, a 0.9375 (e 25000)\n",
      "l: 0.104011, a 0.9512 (e 30000)\n",
      "l: 0.107277, a 0.9570 (e 35000)\n",
      "l: 0.100308, a 0.9668 (e 40000)\n",
      "l: 0.104737, a 0.9863 (e 45000)\n",
      "l: 0.107053, a 0.9766 (e 50000)\n",
      "l: 0.099793, a 0.9961 (e 55000)\n",
      "\n",
      "Final W = \n",
      "\n",
      "[[ 0.57775157 -0.27516143 -0.29983488]\n",
      " [ 0.57768692 -0.27589319 -0.29935118]\n",
      " [ 0.57811469 -0.2764427  -0.29911445]\n",
      " [-0.27712936  0.57835142 -0.29856699]\n",
      " [-0.27640057  0.5782363  -0.29916655]\n",
      " [-0.27706883  0.57939636 -0.29960357]\n",
      " [-0.27661276 -0.27371918  0.55306812]\n",
      " [-0.27658511 -0.27363858  0.55294286]\n",
      " [-0.27656491 -0.2740988   0.55327718]]\n",
      "\n",
      "Final b = \n",
      "\n",
      "[[0.29612778 0.28854414 0.40312817]]\n"
     ]
    }
   ],
   "source": [
    "W, b = train()\n",
    "print('\\nFinal W = \\n\\n{}\\n\\nFinal b = \\n\\n{}'.format(W, b))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "top\n"
     ]
    }
   ],
   "source": [
    "npix = np.array([[255, 143, 255, 255, 187, 93, 5, 5, 5]])\n",
    "a = npix.dot(W) + b\n",
    "d = ['top', 'middle', 'bottom']\n",
    "print(d[np.argmax(a)])"
   ]
  },
  {
   "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": "3.6.10"
  },
  "varInspector": {
   "cols": {
    "lenName": 16,
    "lenType": 16,
    "lenVar": 40
   },
   "kernels_config": {
    "python": {
     "delete_cmd_postfix": "",
     "delete_cmd_prefix": "del ",
     "library": "var_list.py",
     "varRefreshCmd": "print(var_dic_list())"
    },
    "r": {
     "delete_cmd_postfix": ") ",
     "delete_cmd_prefix": "rm(",
     "library": "var_list.r",
     "varRefreshCmd": "cat(var_dic_list()) "
    }
   },
   "types_to_exclude": [
    "module",
    "function",
    "builtin_function_or_method",
    "instance",
    "_Feature"
   ],
   "window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}