Custom KMean.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 152,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.datasets import load_svmlight_files,load_svmlight_file\n",
    "from sklearn.metrics.pairwise import euclidean_distances\n",
    "from copy import deepcopy\n",
    "from random import randint\n",
    "import scipy as sp\n",
    "from scipy import sparse"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "metadata": {},
   "outputs": [],
   "source": [
    "#load the data in SVMLight format\n",
    "iris_data = load_svmlight_file(\"Data/iris.scale\")\n",
    "#create a response decoding dictionary\n",
    "response_dict = {1:'Iris Setosa',2:'Iris Versicolour',3:'Iris Virginica'}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 234,
   "metadata": {},
   "outputs": [],
   "source": [
    "def eucDist(x,y):\n",
    "    '''\n",
    "    measure euclidean distance\n",
    "    '''\n",
    "    diff = x-y\n",
    "    #print(type(diff))\n",
    "    return np.sqrt(diff.multiply(diff).sum(1))\n",
    "\n",
    "def clusterDist(data,centroid):\n",
    "    '''\n",
    "    Distance of a point from cluster\n",
    "    '''\n",
    "    cluster = np.zeros(data.shape[0])\n",
    "    for i in range(data.shape[0]):\n",
    "        cluster[i]=eucDist(data[i],centroid)\n",
    "    return cluster\n",
    "\n",
    "def findCentroid(data,k):\n",
    "    '''\n",
    "    Obtain centroid for k values\n",
    "    Start with random, then take the point distance from all previous centroids\n",
    "    '''\n",
    "    centroids = np.zeros((k,data.shape[1]))\n",
    "    centroids[0] = data[[randint(0,data.shape[0]-1)]].toarray()\n",
    "    dist = np.zeros((data.shape[0],k))\n",
    "    dist[:,0]  = clusterDist(data,sparse.csr_matrix(centroids[0]))\n",
    "    for cls in range(1,k):\n",
    "        centroids[cls] = data[np.argmax(dist.sum(axis=1))].toarray()\n",
    "        dist[:,cls]  = (clusterDist(data,sparse.csr_matrix(centroids[cls])))\n",
    "    #print(centroids)\n",
    "    return centroids\n",
    "\n",
    "def KMeanImproved(data,response,k):\n",
    "    '''\n",
    "    Perform the KMean \n",
    "    '''\n",
    "    cluster_point_new = {}\n",
    "    cluster_point_old = {}\n",
    "    first_pass = True\n",
    "    n= data.shape[0] #number of samples\n",
    "    centroid_old = np.zeros((k,data.shape[1]))# to store old centers\n",
    "    centroid_new = data[list(randint(0,data.shape[0]) for i in range(k))]  #create_centroids(data,k,centroid_new)\n",
    "    clusters = np.zeros(n)\n",
    "    dist = np.zeros((n,k)) #store the distance\n",
    "    while cluster_point_old != cluster_point_new or first_pass:\n",
    "        first_pass = False\n",
    "        for i in range(k):\n",
    "            dist[:,i] = clusterDist(data,centroid_new[i]) #calculate the distance from centroid to all point\n",
    "        clusters = np.array(np.argmin(dist, axis = 1))\n",
    "        cluster_point_old = deepcopy(cluster_point_new)\n",
    "        for i in range(k):\n",
    "            cluster_point_new[i] = list(np.argwhere(clusters == i).flatten()) #assign cluster points\n",
    "            centroid_new[i] = np.array(data[cluster_point_new[i]].mean(axis=0)) #create centroids\n",
    "    cluster_label = build_cluster_label(cluster_point_new,response) #create labels for centroids\n",
    "    return centroid_new,cluster_label,cluster_point_new\n",
    "\n",
    "def KMeanOptimized(data,response,k):\n",
    "    '''\n",
    "    Perform the KMean \n",
    "    '''\n",
    "    cluster_point_new = {}\n",
    "    cluster_point_old = {}\n",
    "    first_pass = True\n",
    "    n= data.shape[0] #number of samples\n",
    "\n",
    "    centroid_new = data[list(randint(0,data.shape[0]) for i in range(k))]  #create_centroids(data,k,centroid_new)\n",
    "    clusters = np.zeros(n)\n",
    "    dist = np.zeros((n,k)) #store the distance\n",
    "    while cluster_point_old != cluster_point_new or first_pass:\n",
    "        first_pass = False\n",
    "        for i in range(k):\n",
    "            dist[:,i] = clusterDist(data,centroid_new[i]) #calculate the distance from centroid to all point\n",
    "        clusters = np.array(np.argmin(dist, axis = 1))\n",
    "        cluster_point_old = deepcopy(cluster_point_new)\n",
    "        for i in range(k):\n",
    "            cluster_point_new[i] = list(np.argwhere(clusters == i).flatten()) #assign cluster points\n",
    "            centroid_new[i] = np.array(data[cluster_point_new[i]].mean(axis=0)) #create centroids\n",
    "    cluster_label = build_cluster_label(cluster_point_new,response) #create labels for centroids\n",
    "    return centroid_new,cluster_label,cluster_point_new\n",
    "\n",
    "def build_cluster_label(cluster_point_new,response):\n",
    "    cluster_label = {}\n",
    "    for index,value in cluster_point_new.items():\n",
    "        #print(type(value))\n",
    "        cluster_label[index] = round(np.sum(response[value])/len(response[value]))\n",
    "    return cluster_label\n",
    "    \n",
    "def predict(data,centroids,clusters,encoding_dict):\n",
    "    '''\n",
    "    Perform prediction \n",
    "    '''\n",
    "    pred_label = []\n",
    "    for row in data:\n",
    "        pred_label.append(clusters[np.argmin([clusterDist(row,sparse.csr_matrix(center)) for center in centroids])])\n",
    "    return pred_label\n",
    "\n",
    "def sse(data,clusters,centroid_new,k):\n",
    "    '''\n",
    "    Evaluate the intra cluster distance\n",
    "    '''\n",
    "    total_error = 0\n",
    "    for i in range(k):\n",
    "        total_error+=np.sum(clusterDist(data[clusters[i]],centroid_new[i]))\n",
    "    return total_error"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Iris Dataset"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### We use elbow method, in which the elbow of the curve between the distortion and cluster value is taken as optimal k value"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 142,
   "metadata": {},
   "outputs": [],
   "source": [
    "#choose the best k value using elbow curve\n",
    "import time\n",
    "scores = {}\n",
    "times = {}\n",
    "#running for 5 different K values\n",
    "for k in range(5):\n",
    "    start = time.time()\n",
    "    centroids,cluster_label,clusters = KMeanImproved(iris_data[0],iris_data[1],k+1)\n",
    "    #print(clusters)\n",
    "    end = time.time()\n",
    "    scores[k+1] = (sse(iris_data[0],clusters,centroids,k+1))\n",
    "    times[k+1]= (end - start)\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 166,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x720 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(20, 10))\n",
    "plt.plot(scores.keys(),scores.values(),'-ro')\n",
    "plt.grid()\n",
    "plt.title(\"Within cluster distance vs K\")\n",
    "plt.xticks(range(1,6))\n",
    "plt.xlabel(\"K\")\n",
    "plt.ylabel(\"Distance\")\n",
    "plt.legend([\"Intra cluster distance\"])\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### We can see the distance decrease and being flattened at around k = 3, hence we can use k as 3"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 165,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x720 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(20, 10))\n",
    "plt.plot(times.keys(),times.values(),'-go')\n",
    "plt.grid()\n",
    "plt.title(\"Time to fit\")\n",
    "plt.xticks(range(1,6))\n",
    "plt.xlabel(\"K\")\n",
    "plt.ylabel(\"Duration\")\n",
    "plt.legend([\"Time for fitting\"])\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Performing a fitting and prediction step on KMeans and it's total time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 235,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Wall time: 2.3 s\n",
      "Wall time: 189 ms\n",
      "Wall time: 48.9 ms\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "58.55770236541426"
      ]
     },
     "execution_count": 235,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "%time centroids,cluster_label,clusters = KMeanImproved(iris_data[0],iris_data[1],3)\n",
    "%time predicted_class  = predict(iris_data[0],centroids,cluster_label,response_dict)\n",
    "%time sse(iris_data[0],clusters,centroids,3)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### The centroids  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 243,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[-0.60777778  0.18166667 -0.84271174 -0.88000014]\n",
      " [-0.11748638 -0.38524591  0.15143093  0.09836071]\n",
      " [ 0.41452981 -0.09829064  0.59408938  0.64957267]]\n"
     ]
    }
   ],
   "source": [
    "print(centroids.toarray())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 246,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 3.0, 2.0, 3.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 3.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 3.0, 2.0, 3.0, 3.0, 3.0, 3.0, 2.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 2.0, 3.0, 3.0, 3.0, 3.0, 3.0, 2.0, 3.0, 2.0, 3.0, 2.0, 3.0, 3.0, 2.0, 2.0, 3.0, 3.0, 3.0, 3.0, 3.0, 2.0, 2.0, 3.0, 3.0, 3.0, 2.0, 3.0, 3.0, 3.0, 2.0, 3.0, 3.0, 3.0, 2.0, 3.0, 3.0, 2.0]\n"
     ]
    }
   ],
   "source": [
    "print(predicted_class)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Cluster news article"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Due to scalability we will create our lightSVM file using the vectorized text corpus and export it to more scalable platforms"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 239,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.feature_extraction.text import TfidfVectorizer\n",
    "from sklearn.datasets import dump_svmlight_file\n",
    "\n",
    "\n",
    "def load_data(data):\n",
    "    '''\n",
    "    loads the data from subdirectory and creates response variable based on the folder name\n",
    "    '''\n",
    "    x = []\n",
    "    y = []\n",
    "    for category in os.listdir(data):\n",
    "        for fileName in os.listdir(data+\"/\"+str(category)):\n",
    "            with open(data+\"/\"+str(category)+\"/\"+str(fileName), 'r') as file:\n",
    "                x.append(file.read())\n",
    "                y.append(category)\n",
    "    return x, y\n",
    "\n",
    "def fetchStopWords():\n",
    "    '''\n",
    "    load the stop words from text file\n",
    "    '''\n",
    "    f = open(\"stopwords.txt\", \"r\")\n",
    "    stopwords = f.read()\n",
    "    return stopwords.split('\\n')\n",
    "\n",
    "def vectorize(train_data,test_data):\n",
    "    '''\n",
    "    Create a TF-IDF and transform both test and train\n",
    "    In real time this will be included as a pipeline component\n",
    "    '''\n",
    "    vectorizer = TfidfVectorizer(stop_words = fetchStopWords(),\n",
    "                                 min_df=10, \n",
    "                                 lowercase = True, \n",
    "                                 ngram_range=(1, 3))\n",
    "    v_train = vectorizer.fit_transform(train_data)\n",
    "    v_test = vectorizer.transform(test_data)\n",
    "    return v_train,v_test\n",
    "\n",
    "def createRespDict(train_response):\n",
    "    '''\n",
    "    Encode the response as numbers\n",
    "    '''\n",
    "    response_dict = {}\n",
    "    index = 1\n",
    "    for i in set(train_response):\n",
    "        response_dict[i] = index\n",
    "        index += 1\n",
    "    return response_dict"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 211,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Wall time: 2min 46s\n",
      "Wall time: 1min 42s\n"
     ]
    }
   ],
   "source": [
    "#load the files into train and test splits\n",
    "%time X_train_words, y_train_words = load_data(\"Data/20news-bydate-train\")\n",
    "%time X_test_words, y_test_words = load_data(\"Data/20news-bydate-test\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 190,
   "metadata": {},
   "outputs": [],
   "source": [
    "#create the TfIDF vectors for the words\n",
    "X_train,X_test = vectorize(X_train_words,X_test_words)\n",
    "#encode the response values to numbers\n",
    "response_dict = createRespDict(y_train_words)\n",
    "#cast the values\n",
    "y_train = [response_dict[x] for x in y_train_words]\n",
    "y_test = [response_dict[x] for x in y_test_words]\n",
    "\n",
    "#create the SVMlight for both train and test data\n",
    "dump_svmlight_file(X_train, y_train, \"train\", zero_based=False,multilabel=False)\n",
    "dump_svmlight_file(X_test, y_test, \"test\", zero_based=False,multilabel=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### We will use this file in the next notebook to produce data"
   ]
  },
  {
   "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": "3.7.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}