0027 SVM Classifier Example on Recognizing Twitter Users.ipynb

{
 "metadata": {
  "name": "",
  "signature": "sha256:b915511a5c2f66cdd055954f9dd2a64129da37e614cecb026472d41edd6719a9"
 },
 "nbformat": 3,
 "nbformat_minor": 0,
 "worksheets": [
  {
   "cells": [
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "%matplotlib inline\n",
      "import sys\n",
      "import string\n",
      "import numpy as np\n",
      "import pandas as pd\n",
      "import matplotlib.pyplot as plt\n",
      "import seaborn as sns\n",
      "\n",
      "from sklearn.svm import SVC\n",
      "from sklearn.metrics import accuracy_score\n",
      "from sklearn.feature_extraction.text import CountVectorizer\n",
      "from sklearn.feature_extraction.text import TfidfTransformer\n",
      "from __future__ import print_function\n",
      "\n",
      "# turn of data table rendering\n",
      "pd.set_option('display.notebook_repr_html', False)\n",
      "sns.set_palette(['#00A99D', '#F5CA0C', '#B6129F', '#76620C', '#095C57'])\n",
      "sys.version"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 1,
       "text": [
        "'3.4.1 |Anaconda 2.1.0 (64-bit)| (default, Sep 24 2014, 18:32:42) [MSC v.1600 64 bit (AMD64)]'"
       ]
      }
     ],
     "prompt_number": 1
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "## Support Vector Machines\n",
      "Support Vector Machines, or SVMs, are supervised learning models used for classification and regression analysis. The concept of how SVMs work can be hard to grasp. It sure is a much more advanced way to classify data than the [Naive Bayes](http://unsupervised-learning.com/gaussian-naive-bayes-classifier-scikit-learn/) algorithm discussed earlier. Given a data set with features and classes, an SVM finds a line with the largest distance between each of two features, called an optimal hyperplane with a maximal margin. It's like a highway through the data points separating the two features as much as possible.\n",
      "\n",
      "The plot shown below, borrowed from the [scikit-learn documentation](http://scikit-learn.org/stable/auto_examples/svm/plot_separating_hyperplane.html#example-svm-plot-separating-hyperplane-py), shows a linear SVM hyperplane on two features with three support vectors in yellow."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "# We create 40 separable points\n",
      "np.random.seed(0)\n",
      "X = np.r_[np.random.randn(20, 2) - [2, 2], np.random.randn(20, 2) + [2, 2]]\n",
      "Y = [0] * 20 + [1] * 20\n",
      "\n",
      "# Fit the model\n",
      "clf = SVC(kernel='linear')\n",
      "clf.fit(X, Y)\n",
      "\n",
      "# Get the separating hyperplane\n",
      "w = clf.coef_[0]\n",
      "a = -w[0] / w[1]\n",
      "xx = np.linspace(-5, 5)\n",
      "yy = a * xx - (clf.intercept_[0]) / w[1]\n",
      "\n",
      "# Plot the parallels to the separating hyperplane that pass through the\n",
      "# support vectors\n",
      "b = clf.support_vectors_[0]\n",
      "yy_down = a * xx + (b[1] - a * b[0])\n",
      "b = clf.support_vectors_[-1]\n",
      "yy_up = a * xx + (b[1] - a * b[0])\n",
      "\n",
      "# Plot the line, the points, and the nearest vectors to the plane\n",
      "plt.plot(xx, yy, 'k-', color='#00A99D')\n",
      "plt.plot(xx, yy_down, 'k--', color='gray')\n",
      "plt.plot(xx, yy_up, 'k--', color='gray')\n",
      "\n",
      "# Plot the support vectors in yellow\n",
      "plt.scatter(clf.support_vectors_[:, 0], clf.support_vectors_[:, 1],\n",
      "            s=200, color='#F5CA0C')\n",
      "plt.scatter(X[:, 0], X[:, 1], c=Y, cmap='gray', s=40)\n",
      "plt.show()"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "display_data",
       "png": "iVBORw0KGgoAAAANSUhEUgAAAeYAAAFVCAYAAAA6zUwUAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n/3VMvJHl1CoEAAAAAElFTkSuQmCC\n",
       "text": [
        "<matplotlib.figure.Figure at 0xa81a470>"
       ]
      }
     ],
     "prompt_number": 2
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "## Kernel functions\n",
      "The example here uses a linear kernel function. Applied to the features it makes the data highly separable by increasing the dimensionality of the data. The scikit-learn SVM implementation had three [built-in kernel functions](http://scikit-learn.org/stable/auto_examples/svm/plot_svm_kernels.html) 'linear', 'poly' and 'rbf'."
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "## Example\n",
      "In this example we use an SVM with a linear kernel to predict who wrote an update on Twitter. The dataset contains recent tweets of two data analytics giants; Gregory Piatetsky ([@kdnuggets](https://twitter.com/kdnuggets)) and Big Data Science ([@analyticbridge](https://twitter.com/analyticbridge)). The screen_name acts as the label while each word in the tweet is a feature.\n",
      "\n",
      "You can [download the data set](https://raw.githubusercontent.com/remondo/NoteBooks-Unsupervised-Learning/master/data/tweets.csv) from Github."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "# Load 200 tweets from @kdnuggets and @analyticbridge (n=400)\n",
      "df = pd.read_csv('data/tweets.csv')\n",
      "df.screen_name.value_counts()"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 3,
       "text": [
        "kdnuggets         200\n",
        "analyticbridge    200\n",
        "dtype: int64"
       ]
      }
     ],
     "prompt_number": 3
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "# Show arbitrary datapoint\n",
      "df.screen_name[331], df.tweet[331]"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 4,
       "text": [
        "('analyticbridge',\n",
        " 'Three myths about data scientists and big data http://t.co/EtzjR4pdbH')"
       ]
      }
     ],
     "prompt_number": 4
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "# Shuffle the data\n",
      "def shuffle(data, n):\n",
      "    ind = data.index\n",
      "    for i in range(n):\n",
      "       sampler = np.random.permutation(data.shape[0])\n",
      "       new_vals = data.take(sampler).values\n",
      "       data = pd.DataFrame(new_vals, index=ind)\n",
      "    return data\n",
      "\n",
      "df = shuffle(df, 1)\n",
      "df.columns = ['screen_name', 'tweet']\n",
      "df.head()"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 5,
       "text": [
        "      screen_name                                              tweet\n",
        "0  analyticbridge  Six keywords characterizing milestones in the ...\n",
        "1       kdnuggets  Get practical training at PASS Business Analyt...\n",
        "2  analyticbridge  The application of Propensity Score Matching h...\n",
        "3       kdnuggets  Old data analysts don't die - they just get br...\n",
        "4       kdnuggets  How Many Quants are Changing Jobs? http://t.co..."
       ]
      }
     ],
     "prompt_number": 5
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "## Training and test sets\n",
      "To train our SVM model, we need a large enough training set and a test set to test for accuracy. We split the data 75% for training and 25% for testing the model."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "# Create a training set (3/4 of the data)\n",
      "train_n = 300\n",
      "train_features = df['tweet'][:train_n]\n",
      "train_labels = df['screen_name'][:train_n]\n",
      "train_features.count()"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 6,
       "text": [
        "300"
       ]
      }
     ],
     "prompt_number": 6
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "# Create a test set (1/4 of the data)\n",
      "test_features = df['tweet'][train_n:]\n",
      "test_labels = df['screen_name'][train_n:]\n",
      "test_features.count()"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 7,
       "text": [
        "100"
       ]
      }
     ],
     "prompt_number": 7
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "## TF-IDF Feature Vectors\n",
      "To extract useful features from the tweets we use the TF-IDF algorithm to qualify the importance of words. For more on TF-IDF, see [Using TF-IDF for Text Mining in Python](http://unsupervised-learning.com/tf-idf-text-mining-python). To do this, we use the [count_vectorizer](http://scikit-learn.org/stable/modules/generated/sklearn.feature_extraction.text.CountVectorizer.html) and [tfidf_transformer](http://scikit-learn.org/stable/modules/generated/sklearn.feature_extraction.text.TfidfTransformer.html) from the scikit-learn machine learning library."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "# In order to perform machine learning on text documents, we first \n",
      "# need to turn the text content into numerical features.\n",
      "count_vectorizer = CountVectorizer()\n",
      "\n",
      "train_counts = count_vectorizer.fit_transform(train_features)\n",
      "train_counts.shape"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 8,
       "text": [
        "(300, 1688)"
       ]
      }
     ],
     "prompt_number": 8
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "# Transform to Term Frequencies - Inverse Document Frequencies\n",
      "tfidf_transformer = TfidfTransformer(use_idf=False)\n",
      "tfidf_transformer.fit(train_counts)\n",
      "\n",
      "train_tfidf = tfidf_transformer.transform(train_counts)\n",
      "train_tfidf.shape"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 9,
       "text": [
        "(300, 1688)"
       ]
      }
     ],
     "prompt_number": 9
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "# Create a TF-IDF for the test data\n",
      "test_counts = count_vectorizer.transform(test_features)\n",
      "test_tfidf = tfidf_transformer.transform(test_counts)\n",
      "test_tfidf.shape"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 10,
       "text": [
        "(100, 1688)"
       ]
      }
     ],
     "prompt_number": 10
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "## Training the SVM Classifier\n",
      "We train the Support Vector Machine with a linear kernel on the TF-IDF scores and both Twitter handles as the training labels."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "# Train the SVC with linear kernel\n",
      "clf = SVC(kernel=\"linear\")\n",
      "clf.fit(train_tfidf, train_labels)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 11,
       "text": [
        "SVC(C=1.0, cache_size=200, class_weight=None, coef0=0.0, degree=3, gamma=0.0,\n",
        "  kernel='linear', max_iter=-1, probability=False, random_state=None,\n",
        "  shrinking=True, tol=0.001, verbose=False)"
       ]
      }
     ],
     "prompt_number": 11
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "## Prediction\n",
      "To test our fitted model, we use the test set to predict tweets it has never seen before."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "# Predict with the test data\n",
      "pred = clf.predict(test_tfidf)\n",
      "pred[:11]"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 12,
       "text": [
        "array(['kdnuggets', 'analyticbridge', 'analyticbridge', 'analyticbridge',\n",
        "       'analyticbridge', 'kdnuggets', 'analyticbridge', 'analyticbridge',\n",
        "       'kdnuggets', 'kdnuggets', 'kdnuggets'], dtype=object)"
       ]
      }
     ],
     "prompt_number": 12
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "## Model Accuracy\n",
      "We determine the current model accuracy by comparing the actual test labels with the predicted labels."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "# Plot the accuracy of the model\n",
      "sns.barplot(test_labels == pred, color='#00A99D')\n",
      "plt.show()"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "display_data",
       "png": "iVBORw0KGgoAAAANSUhEUgAAAe0AAAFkCAYAAAANC2PrAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nbUmSCmG0JUkqxP8D15OUrB+oCfcAAAAASUVORK5CYII=\n",
       "text": [
        "<matplotlib.figure.Figure at 0xbf12b00>"
       ]
      }
     ],
     "prompt_number": 13
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "# Calculate the accuracy\n",
      "accuracy = accuracy_score(pred, test_labels)\n",
      "accuracy"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 14,
       "text": [
        "0.71999999999999997"
       ]
      }
     ],
     "prompt_number": 14
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "So it works, but the model is not accurate enough. Maybe not that bad since the model is trained on a relatively small dataset."
     ]
    }
   ],
   "metadata": {}
  }
 ]
}