Notebooks/Spark-Example-19b-SVM.ipynb

{
 "cells": [
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# SVM Implementation\n",
    "Author: Uwe Mueller, June, 2023 v1.0\n",
    "\n",
    "Implementation of SVM with a weighted stochastic gradient descent algorithm."
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Loss Function\n",
    "\n",
    "$$\n",
    "\\frac{\\lambda}{2} || w ||^2 + \\frac{1}{n} \\sum\\limits_i max(0, 1 - y_i (w.x_i)) \\qquad\\qquad{(4.9)}\n",
    "$$"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Gradient \n",
    "$$\n",
    "\n",
    "\\\n",
    "\\begin{align*}\n",
    "\\nabla J(w) = \\begin{cases}\n",
    "-w + C \\sum_{i=1}^{m} y^{(i)}x^{(i)}, & \\text{if } y^{(i)}w^Tx^{(i)} < 1 \\\\\n",
    "-w, & \\text{otherwise}\n",
    "\\end{cases}\n",
    "\\end{align*}\n",
    "\\\n",
    "$$"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from __future__ import print_function\n",
    "import re\n",
    "import sys\n",
    "import numpy as np\n",
    "from operator import add\n",
    "import findspark\n",
    "findspark.init()\n",
    "from pyspark import SparkContext\n",
    "from pyspark.sql import SparkSession\n",
    "from pyspark import SparkConf,SparkContext\n",
    "from sklearn.datasets import make_blobs\n",
    "from pandas import DataFrame\n",
    "from matplotlib import pyplot\n",
    "import matplotlib.cm as cm\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt \n",
    "\n",
    "sc = SparkContext(master='local',appName=\"SVM\")\n",
    "spark = SparkSession(sparkContext=sc)\n",
    "spark.sparkContext.setLogLevel('WARN')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Implemeted weighted SVM function with stocastic gradient and regularization\n",
    "def SVM_reg_weighted(data, learningRate, num_Iter, regularisation = 0, printcoeff = False):\n",
    "\t# Calculate the dimensionality of the inpit data\n",
    "    num_samples = data.count()\n",
    "    beta = 0.9\n",
    "    d_input = len(data.first()[1])\n",
    "    qti = coeff = nzeros = np.zeros(d_input)\n",
    "    epsilon = 10e-12\n",
    "    cost_list = []\n",
    "        \n",
    "\n",
    "\t# The main iterative part of gradient descent algorithm \n",
    "    for i in range(num_Iter):\n",
    "\t\t# create mini batch\n",
    "        mini_batch = data.sample(False, 800, i)\n",
    "\t\t# Seperately calculate the gradient for class 0 and class 1\n",
    "        res0 = (mini_batch\n",
    "                .filter(lambda x: x[0] == -1)\n",
    "                .map(lambda x: (x[0], x[1], x[0]*np.dot(x[1], coeff)))\n",
    "                .map(lambda x: (nzeros if x[2] > 1.0 else -1.0 * x[0] * x[1],\\\n",
    "                                max(0, 1 - x[2])))\n",
    "                .reduce(lambda x, y: (x[0] + y[0], x[1] + y[1]))\n",
    "        )\n",
    "\t\n",
    "        res1 =  (mini_batch\n",
    "                .filter(lambda x: x[0] == 1)\n",
    "                .map(lambda x: (x[0], x[1], x[0]*np.dot(x[1], coeff)))\n",
    "                .map(lambda x: (nzeros if x[2] > 1.0 else -1.0 * x[0] * x[1],\\\n",
    "                                max(0, 1 - x[2])))\n",
    "                .reduce(lambda x, y: (x[0] + y[0], x[1] + y[1]))\n",
    "        )\n",
    "\n",
    "\t# Calculate the combined weighted gradient for class 1 and 0\n",
    "        gradients = w0 * res0[0] + w1 * res1[0]\n",
    "        sum_cost = w0 * res0[1] + w1 * res1[1]\t\n",
    "\t\n",
    "\t\n",
    "\t# Add L2 Regularization (sum of squared coefficients)\n",
    "        cost=(sum_cost/float(num_samples))+regularisation*np.square(coeff).sum()\n",
    "        gradient= (1.0/float(num_samples))*gradients+regularisation*2*coeff\n",
    "        sq_grad = gradient ** 2\n",
    "        exp_grad = beta * qti / (i + 1) + (1 - beta) * sq_grad\n",
    "        coeff = coeff - learningRate / np.sqrt(exp_grad + epsilon) * gradient\n",
    "        qti += sq_grad\n",
    "\n",
    "        if printcoeff:\n",
    "            print(f\"{i} Cost:{cost:.5f} Coeff {coeff}\")\n",
    "\t\t# Save the model parameters so we can eveluate the model on the test dataset.\n",
    "\n",
    "\t\t#append cost so we can plot it later\n",
    "        cost_list.append(cost)\n",
    "\n",
    "\t\t# Check if cost is increasing at which point set conversion to False\n",
    "        if i == 0:\n",
    "            start_cost = cost\n",
    "\n",
    "    return(cost_list, coeff, start_cost > cost )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Calculate the Model performance based on the predicition coefficient and the test data and return accuracy, and f1 scores\n",
    "def calc_modelperf(pred_coeff, test_data, printing = False):\n",
    "    num_samples = test_data.count()\n",
    "    predictions = test_data.map(lambda x: (x[0], 1 if np.dot(x[1],pred_coeff)>0 else -1))\n",
    "    TP= predictions.map(lambda x: 1 if (x[0]== 1) and (x[1]==1) else 0).reduce(lambda x,y:x+y)\n",
    "    FP = predictions.map(lambda x: 1 if (x[0]== -1) and (x[1]==1) else 0).reduce(lambda x,y:x+y)\n",
    "\n",
    "    TN = predictions.map(lambda x: 1 if (x[0]== -1) and (x[1]==-1) else 0).reduce(lambda x,y:x+y)\n",
    "    FN = predictions.map(lambda x: 1 if (x[0]== 1) and (x[1]==-1) else 0).reduce(lambda x,y:x+y)\n",
    "\n",
    "    # Calculate the Accuracy and the F1\n",
    "    accuracy = (TP+TN)/(num_samples)\n",
    "    f1 = (2*TP)/((2*TP) + FP + FN)\n",
    "    if TP + FP ==0:\n",
    "          precision = 0\n",
    "    else:\n",
    "          precision = TP / (TP + FP)\n",
    "    if TP + FN == 0:\n",
    "          recall = 0\n",
    "    else:\n",
    "          recall = TP / (TP + FN)\n",
    "\n",
    "    # Print the Contingency matrix\n",
    "    if printing: \n",
    "        print(\"--Contingency matrix--\")\n",
    "        print(f\" TP:{TP:6}  FP:{FP:6}\")\n",
    "        print(f\" FN:{FN:6}  TN:{TN:6}\")\n",
    "        print(\"----------------------\")\n",
    "        print(f\"Accuracy = {accuracy:.4g}  \\nF1 = {f1:.4g} \\nPrecision = {precision:.4g}\\nRecall = {recall:.4g}\")\n",
    "    \n",
    "    return (accuracy, f1)\n"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Create Training/Test Data Set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "n_feature = 2 # the number of features\n",
    "n_components = 2 # the number of clusters\n",
    "\n",
    "max_iter = 4 # max iterations \n",
    "n=100000 # the number of total samples\n",
    "\n",
    "ratio_positive = 0.5 # The classes ratio.\n",
    "n_samples = [int(n*(1-ratio_positive)), int(n*ratio_positive)]\n",
    "n_components = [1,1]\n",
    "n_samples\n",
    "\n",
    "# generate isotropic Gaussian blobs (data from mixture Gaussian distributions)\n",
    "X, y = make_blobs(n_samples=n_samples, # the number of total samples\n",
    "                  centers=None, # the number of clusters\n",
    "                  n_features=n_feature, # the number of features\n",
    "                  cluster_std = 1.2, # the std of each cluster\n",
    "                  random_state=2) # 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# We need to fix the labels so that the labels are -1 and 1 as SVM expects them\n",
    "y = [-1 if x == 0 else 1 for x in y]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Code for plotting the data\n",
    "df = DataFrame(dict(x=X[:,0], y=X[:,1], label=y))\n",
    "cluster_name = set(y)\n",
    "colors = dict(zip(cluster_name, cm.rainbow(np.linspace(0, 1, len(cluster_name)))))\n",
    "fig, ax = pyplot.subplots()\n",
    "grouped = df.groupby('label')\n",
    "for key, group in grouped:\n",
    "    group.plot(ax=ax, kind='scatter', x='x', y='y', label=key, color=colors[key].reshape(1,-1))\n",
    "pyplot.title('Original 2D Data from {} Clusters'.format(n_components))\n",
    "pyplot.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(n of 1, 0 and total): (39988, 40037, 80025)\n",
      "Weights class 0 and 1: 0.9993880660389141 1.0006126838051415\n"
     ]
    }
   ],
   "source": [
    "# read X, y into rdd\n",
    "rdd_X = sc.parallelize(X)\n",
    "rdd_y = sc.parallelize(y)\n",
    "\n",
    "# split data into train and test\n",
    "# Add: map(lambda x: (x[0],np.append(x[1],1)))\n",
    "traindata, testdata = rdd_y.zip(rdd_X).randomSplit([0.8, 0.2], seed=12345)\n",
    "traindata = traindata.map(lambda x: (x[0],np.append(x[1],1)))\n",
    "testdata  = testdata.map(lambda x: (x[0],np.append(x[1],1)))\n",
    "traindata.cache()\n",
    "train_size = traindata.count()\n",
    "t = (traindata\n",
    "    .map(lambda x: (0 if x[0] == -1 else 1,0 if x[0] == 1 else 1, 1))\n",
    "    .reduce(lambda x, y: (x[0] + y[0], x[1] + y[1], x[2] + y[2]))\n",
    "\t)\n",
    "\n",
    "# Check balance of the data and set the weights correctly\n",
    "print(\"(n of 1, 0 and total):\", t)\n",
    "w0 = t[2] / (2*t[1])\n",
    "w1 = t[2] / (2*t[0])\n",
    "print(\"Weights class 0 and 1:\", w0,w1)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 Cost:1.00000 Coeff [0.0284605 0.0284605 0.       ]\n",
      "1 Cost:0.85112 Coeff [0.04059511 0.04059582 0.        ]\n",
      "2 Cost:0.78764 Coeff [0.05135131 0.05135264 0.        ]\n",
      "3 Cost:0.73138 Coeff [0.06157374 0.06157571 0.        ]\n",
      "4 Cost:0.67791 Coeff [0.07148758 0.07145712 0.02844799]\n",
      "5 Cost:0.62655 Coeff [0.08057342 0.07937689 0.05687738]\n",
      "6 Cost:0.58558 Coeff [0.08735883 0.08034954 0.08424734]\n",
      "7 Cost:0.57399 Coeff [0.09412806 0.08134017 0.10344761]\n",
      "8 Cost:0.56267 Coeff [0.10049925 0.08121968 0.12078902]\n",
      "9 Cost:0.55592 Coeff [0.10694283 0.08136578 0.13589993]\n"
     ]
    }
   ],
   "source": [
    "# Train the SVM\n",
    "rescost, rescoeff, _ = SVM_reg_weighted(traindata, learningRate = 0.009, num_Iter = 10, regularisation=0.002, printcoeff = True) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "--Contingency matrix--\n",
      " TP:  9626  FP:   221\n",
      " FN:   386  TN:  9742\n",
      "----------------------\n",
      "Accuracy = 0.9696  \n",
      "F1 = 0.9694 \n",
      "Precision = 0.9776\n",
      "Recall = 0.9614\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(0.9696120150187735, 0.9694345133188982)"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Display Confusion Matrix and some comoon performance metrics\n",
    "calc_modelperf(rescoeff, testdata, printing=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "spark.stop()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "spark",
   "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": "3.10.14"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}