Pyro code for Chapter 3.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import torch\n",
    "import pyro\n",
    "import pyro.distributions as dist\n",
    "import pyro.infer\n",
    "import pyro.optim\n",
    "from pyro.optim import Adam\n",
    "import torch.distributions.constraints as constraints\n",
    "import scipy\n",
    "import scipy.stats\n",
    "\n",
    "\n",
    "def scale(Data):\n",
    "    Jskill = pyro.sample('JS', dist.Normal(120.,40.))\n",
    "    Jperf = pyro.sample('JP', dist.Normal(Jskill, 5.))\n",
    "    Fskill = pyro.sample('FS', dist.Normal(100.,5.)) \n",
    "    Fperf = pyro.sample('FP', dist.Normal(Fskill, 5.))\n",
    "    Jwin = torch.tensor([0.0])\n",
    "    if Fperf<Jperf:\n",
    "        Jwin = torch.tensor([1.0])\n",
    "    with pyro.plate(\"plate\", len(Data)):\n",
    "        a= pyro.sample('JW', dist.Bernoulli(Jwin), obs=Data)\n",
    "\n",
    "def guide(Data):\n",
    "    J_m = pyro.param('J_m', torch.tensor(0.0))\n",
    "    J_s = pyro.param('J_s', torch.tensor(5.0), constraint = constraints.positive)\n",
    "    F_m = pyro.param('F_m', torch.tensor(0.0))\n",
    "    F_s = pyro.param('F_s', torch.tensor(5.0), constraint=constraints.positive)\n",
    "    Jp_s = pyro.param('JP_s', torch.tensor(25.0), constraint=constraints.positive)\n",
    "    Fp_s = pyro.param('FP_s', torch.tensor(25.0),constraint=constraints.positive)\n",
    "    Jskill = pyro.sample('JS', dist.Normal(J_m,J_s))\n",
    "    Jperf = pyro.sample('JP', dist.Normal(Jskill, Jp_s))\n",
    "    Fskill = pyro.sample('FS', dist.Normal(F_m,F_s))\n",
    "    Fperf = pyro.sample('FP', dist.Normal(Fskill,Fp_s))\n",
    "    \n",
    "Data =  torch.tensor([1.0])\n",
    "\n",
    "pyro.clear_param_store()\n",
    "adam_params = {\"lr\": 0.09, \"betas\": (.9, .999)}\n",
    "optimizer = Adam(adam_params)\n",
    "svi = pyro.infer.SVI(model=scale,\n",
    "                     guide=guide,\n",
    "                     optim=optimizer,\n",
    "                     loss=pyro.infer.TraceGraph_ELBO())\n",
    "\n",
    "\n",
    "losses, a,b,c,d  = [], [], [], [], []\n",
    "num_steps = 5000\n",
    "for t in range(num_steps):\n",
    "    losses.append(svi.step(Data))\n",
    "    a.append(pyro.param(\"J_s\").item())\n",
    "    b.append(pyro.param(\"J_m\").item())\n",
    "    c.append(pyro.param(\"F_s\").item())\n",
    "    d.append(pyro.param(\"F_m\").item())\n",
    "    if t%500 ==0: \n",
    "        print('.')\n",
    "\n",
    "plt.subplot(5,1,1)\n",
    "plt.plot(losses)\n",
    "plt.title(\"Adam\")\n",
    "plt.xlabel(\"step\")\n",
    "plt.ylabel(\"loss\");\n",
    "\n",
    "plt.subplot(5,1,2)\n",
    "plt.plot(a)\n",
    "plt.xlabel('step')\n",
    "plt.ylabel('Js')\n",
    "\n",
    "plt.subplot(5,1,3)\n",
    "plt.plot(b)\n",
    "plt.xlabel('step')\n",
    "plt.ylabel('Jm')\n",
    "\n",
    "plt.subplot(5,1,4)\n",
    "plt.plot(c)\n",
    "plt.xlabel('step')\n",
    "plt.ylabel('Fs')\n",
    "\n",
    "plt.subplot(5,1,5)\n",
    "plt.plot(d)\n",
    "plt.xlabel('step')\n",
    "plt.ylabel('Fm')\n",
    "\n",
    "plt.savefig('loss.png')\n",
    "print('Jskills = N({1},{0} ), Fskills=N({3},{2}) '.format(pyro.param(\"J_s\").item(), pyro.param(\"J_m\").item(), pyro.param(\"F_s\").item(), pyro.param(\"F_m\").item()))\n",
    "\n",
    "\n",
    "print(\"the book obtains the solution N(140.1, 28.5), we obtained {}\".format(pyro.param(\"J_m\").item()))\n",
    "print((140.1- pyro.param(\"J_m\").item())/28.5, ' standar devations from the book solution.')\n"
   ]
  }
 ],
 "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": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}