delta_rule_chap12_Anderson

delta_rule_chap12_anderson.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "delta_rule_chap12_Anderson",
      "provenance": [],
      "collapsed_sections": [],
      "authorship_tag": "ABX9TyNLS5HFxuNAQb8yJumAzWM5",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/gist/sigmaroles/b7b7955e8254260dbe25389cadc164b9/delta_rule_chap12_anderson.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ca1om2FOGLGF"
      },
      "source": [
        "**Anderson chapter 12:** <br>\n",
        "12.2 Octave Functions for the Delta Rule Exercise <br>\n",
        "and <br>\n",
        "12.3 Using Octave to Solve the Delta Rule Exercise"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "QoZJFWjS1zWv"
      },
      "source": [
        "import numpy as np\n",
        "rand = np.random.rand"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4yy6JYAv0sIc"
      },
      "source": [
        "def mkLine():\n",
        "  return rand(), rand()\n",
        "\n",
        "def mkXs(n):\n",
        "  return rand(n,1)*40 - 20"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "LjOqTGe-2p8F"
      },
      "source": [
        "def mkYs(m,b,xs):\n",
        "  #xs = np.array([1,2,3,4]).reshape(-1,1)\n",
        "  #m, b = mkLine()\n",
        "  ys = m * xs + b\n",
        "  ys = ys.reshape(-1,1)\n",
        "\n",
        "  c1 = np.tile([1, -1], (round(len(xs)/2), 1)).reshape(-1,1)\n",
        "  # repmat is tile in numpy: https://docs.scipy.org/doc/numpy-1.15.0/user/numpy-for-matlab-users.html\n",
        "  # reshape is to make it one-dimensional\n",
        "\n",
        "  yp = ys + np.multiply(5*rand(len(xs),1), c1)\n",
        "  temp1 = np.tile([1], (len(xs),1))\n",
        "  ymat = np.hstack([c1,ys,xs,yp,temp1])\n",
        "\n",
        "  return ymat"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BAHLuEbr7l4S"
      },
      "source": [
        "def compOut(t,wv,iv):\n",
        "  a = np.dot(wv,iv);\n",
        "  if a>=t:\n",
        "    return 1\n",
        "  else:\n",
        "    return -1"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2M6PKmaW9Hkq"
      },
      "source": [
        "def dR(eta, obs, des, iv, wv):\n",
        "  return eta * np.dot((des-obs),iv) + wv"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "n1MY94L19SeQ"
      },
      "source": [
        "def oneLoop(t,eta, cls, iv, wv):\n",
        "  obs = compOut(t,wv,iv)\n",
        "  nwv = dR(eta, obs, cls, iv, wv)\n",
        "  return nwv"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "6v166GWp9gcK"
      },
      "source": [
        "def onePass(dataMat, wv, eta, t):\n",
        "  ss = dataMat.shape[0]\n",
        "  for i in range(ss):\n",
        "    wv = oneLoop(t,eta,dataMat[i,0], [dataMat[i,2], dataMat[i,3], dataMat[i,4]], wv)\n",
        "  return wv"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "G4BxJRD4BwmL"
      },
      "source": [
        "def train(dataMat, wv, eta, t):\n",
        "  epsilon = 0.001\n",
        "  owv = [0, 0, 0]\n",
        "  nwv = wv\n",
        "  while (np.sum(np.abs(nwv-owv)) > epsilon):\n",
        "    owv = nwv\n",
        "    nwv = onePass(dataMat, nwv, eta, t)\n",
        "  \n",
        "  return nwv"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "k_VanL-KCOaj"
      },
      "source": [
        "s, i = mkLine()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2NYEa4aeClRc"
      },
      "source": [
        "xs = mkXs(20)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0cBdO6JFCnOz"
      },
      "source": [
        "dm = mkYs(s,i,xs)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1MRDrSESCpiD"
      },
      "source": [
        "trainSet = dm[:10, :]\n",
        "testSet = dm[10:20, :]"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "sgjhLR1kC1mP"
      },
      "source": [
        "saveWt = train(trainSet, rand(1,3), 0.1, 0)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "zypn9WHkDT-P",
        "outputId": "c0211b12-ab03-4c1d-ddf8-36a975904972"
      },
      "source": [
        "# indices must be reduced by one because Python arrays start at zero, Matlab's start at one\n",
        "(np.dot(saveWt, trainSet[:,[2,3,4]].transpose()) > 0) * 2 - 1"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([[ 1, -1,  1, -1,  1, -1,  1, -1,  1, -1]])"
            ]
          },
          "metadata": {},
          "execution_count": 14
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "-Kj4K3b8E4FY",
        "outputId": "e65e9e0c-1afd-46f4-95a3-ac8aa2e3103d"
      },
      "source": [
        "# what about test set?\n",
        "(np.dot(saveWt, testSet[:,[2,3,4]].transpose()) > 0) * 2 - 1"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([[ 1, -1,  1, -1,  1, -1,  1, -1,  1, -1]])"
            ]
          },
          "metadata": {},
          "execution_count": 15
        }
      ]
    }
  ]
}