dual-gradient-descent.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "from lib import DualGradientDescent\n",
    "from lib import Experiments"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true,
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "def sample_batch(n):\n",
    "    c = np.random.uniform(0, 10, size=[n, n])\n",
    "    p = np.random.dirichlet(np.ones(n), size=1).ravel()\n",
    "    q = np.random.dirichlet(np.ones(n), size=1).ravel()\n",
    "    return c, p, q"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "iteration 5000:   criteria 1 = 0.0256658, criteria 2 = -0.2474279\n",
      "iteration 10000:   criteria 1 = 0.0141456, criteria 2 = -0.1885392\n",
      "iteration 15000:   criteria 1 = 0.0096324, criteria 2 = -0.1617641\n",
      "iteration 20000:   criteria 1 = 0.0072496, criteria 2 = -0.1469809\n",
      "iteration 25000:   criteria 1 = 0.0058073, criteria 2 = -0.1379122\n",
      "iteration 30000:   criteria 1 = 0.004845, criteria 2 = -0.1317915\n",
      "iteration 35000:   criteria 1 = 0.0041555, criteria 2 = -0.1273553\n",
      "iteration 40000:   criteria 1 = 0.0036366, criteria 2 = -0.1239978\n",
      "iteration 45000:   criteria 1 = 0.0032326, criteria 2 = -0.1213742\n"
     ]
    }
   ],
   "source": [
    "n = 40\n",
    "c, p, q = sample_batch(n)\n",
    "x, k = DualGradientDescent(gamma=0.05, epsilon=1e-2, n=n).fit(c, p, q)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x1a13c12f60>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "c_, p_, q_ = Experiments.load_data('experiments/test_data/1.png', 'experiments/test_data/2.png', show=True, size=10)\n",
    "n = p_.shape[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "iteration 5000:   criteria 1 = 0.0265669, criteria 2 = -0.2716451\n",
      "iteration 10000:   criteria 1 = 0.0147016, criteria 2 = -0.2051038\n",
      "iteration 15000:   criteria 1 = 0.0103347, criteria 2 = -0.177832\n",
      "iteration 20000:   criteria 1 = 0.0080651, criteria 2 = -0.1630663\n",
      "iteration 25000:   criteria 1 = 0.0066411, criteria 2 = -0.1533913\n",
      "iteration 30000:   criteria 1 = 0.0056758, criteria 2 = -0.1467923\n",
      "iteration 35000:   criteria 1 = 0.0049803, criteria 2 = -0.1420276\n",
      "iteration 40000:   criteria 1 = 0.0044512, criteria 2 = -0.1383686\n",
      "iteration 45000:   criteria 1 = 0.0040357, criteria 2 = -0.1354954\n",
      "iteration 50000:   criteria 1 = 0.0037016, criteria 2 = -0.133188\n",
      "iteration 55000:   criteria 1 = 0.0034261, criteria 2 = -0.1312765\n",
      "iteration 60000:   criteria 1 = 0.0031941, criteria 2 = -0.1296488\n",
      "iteration 65000:   criteria 1 = 0.0029947, criteria 2 = -0.1282212\n",
      "iteration 70000:   criteria 1 = 0.0028203, criteria 2 = -0.1269392\n",
      "iteration 75000:   criteria 1 = 0.0026663, criteria 2 = -0.1257871\n"
     ]
    }
   ],
   "source": [
    "x, k = DualGradientDescent(gamma=0.05, epsilon=1e-2, n=n).fit(c_, p_, q_)"
   ]
  },
  {
   "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
}