Nested versus non-nested cross-validation.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Nested versus non-nested cross-validation"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Imports"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.datasets import load_iris\n",
    "from matplotlib import pyplot as plt\n",
    "from sklearn.svm import SVC\n",
    "from sklearn.model_selection import GridSearchCV, cross_val_score, KFold\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Main Program"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "iteration: 1 | best parameters: {'C': 1, 'gamma': 0.1}\n",
      "iteration: 2 | best parameters: {'C': 10, 'gamma': 0.1}\n",
      "iteration: 3 | best parameters: {'C': 100, 'gamma': 0.01}\n",
      "iteration: 4 | best parameters: {'C': 10, 'gamma': 0.01}\n",
      "iteration: 5 | best parameters: {'C': 10, 'gamma': 0.01}\n",
      "iteration: 6 | best parameters: {'C': 10, 'gamma': 0.1}\n",
      "iteration: 7 | best parameters: {'C': 100, 'gamma': 0.01}\n",
      "iteration: 8 | best parameters: {'C': 100, 'gamma': 0.01}\n",
      "iteration: 9 | best parameters: {'C': 10, 'gamma': 0.1}\n",
      "iteration: 10 | best parameters: {'C': 100, 'gamma': 0.01}\n",
      "iteration: 11 | best parameters: {'C': 100, 'gamma': 0.01}\n",
      "iteration: 12 | best parameters: {'C': 10, 'gamma': 0.01}\n",
      "iteration: 13 | best parameters: {'C': 1, 'gamma': 0.1}\n",
      "iteration: 14 | best parameters: {'C': 10, 'gamma': 0.01}\n",
      "iteration: 15 | best parameters: {'C': 100, 'gamma': 0.01}\n",
      "iteration: 16 | best parameters: {'C': 1, 'gamma': 0.1}\n",
      "iteration: 17 | best parameters: {'C': 10, 'gamma': 0.1}\n",
      "iteration: 18 | best parameters: {'C': 100, 'gamma': 0.01}\n",
      "iteration: 19 | best parameters: {'C': 1, 'gamma': 0.1}\n",
      "iteration: 20 | best parameters: {'C': 100, 'gamma': 0.01}\n",
      "iteration: 21 | best parameters: {'C': 100, 'gamma': 0.01}\n",
      "iteration: 22 | best parameters: {'C': 100, 'gamma': 0.01}\n",
      "iteration: 23 | best parameters: {'C': 10, 'gamma': 0.1}\n",
      "iteration: 24 | best parameters: {'C': 10, 'gamma': 0.01}\n",
      "iteration: 25 | best parameters: {'C': 10, 'gamma': 0.1}\n",
      "iteration: 26 | best parameters: {'C': 100, 'gamma': 0.01}\n",
      "iteration: 27 | best parameters: {'C': 1, 'gamma': 0.1}\n",
      "iteration: 28 | best parameters: {'C': 1, 'gamma': 0.1}\n",
      "iteration: 29 | best parameters: {'C': 100, 'gamma': 0.01}\n",
      "iteration: 30 | best parameters: {'C': 10, 'gamma': 0.1}\n"
     ]
    }
   ],
   "source": [
    "# Number of random trials\n",
    "NUM_TRIALS = 30\n",
    "\n",
    "# Load the dataset\n",
    "iris = load_iris()\n",
    "X_iris = iris.data\n",
    "y_iris = iris.target\n",
    "\n",
    "# Set up possible values of parameters to optimize over\n",
    "p_grid = {\"C\": [1, 10, 100],\n",
    "          \"gamma\": [.01, .1]}\n",
    "\n",
    "# We will use a Support Vector Classifier with \"rbf\" kernel\n",
    "svm = SVC(kernel=\"rbf\")\n",
    "\n",
    "# Arrays to store scores\n",
    "non_nested_scores = np.zeros(NUM_TRIALS)\n",
    "nested_scores = np.zeros(NUM_TRIALS)\n",
    "\n",
    "# Loop for each trial\n",
    "for i in range(NUM_TRIALS):\n",
    "    # Choose cross-validation techniques for the inner and outer loops,\n",
    "    # independently of the dataset.\n",
    "    # E.g \"GroupKFold\", \"LeaveOneOut\", \"LeaveOneGroupOut\", etc.\n",
    "    inner_cv = KFold(n_splits=4, shuffle=True, random_state=i)\n",
    "    outer_cv = KFold(n_splits=4, shuffle=True, random_state=i)\n",
    "\n",
    "    # Non_nested parameter search and scoring\n",
    "    clf = GridSearchCV(estimator=svm, param_grid=p_grid, cv=inner_cv)\n",
    "    # fit using best params\n",
    "    clf.fit(X_iris, y_iris)\n",
    "    non_nested_scores[i] = clf.best_score_\n",
    "    \n",
    "    print ('iteration: {} | best parameters: {}'.format(i+1, clf.best_params_))\n",
    "\n",
    "    # Nested CV with parameter optimization\n",
    "    # unbiased estimation of validation accuracy\n",
    "    nested_score = cross_val_score(clf, X=X_iris, y=y_iris, cv=outer_cv)\n",
    "    nested_scores[i] = nested_score.mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Average difference of 0.007742 with std. dev. of 0.007688.\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "score_difference = non_nested_scores - nested_scores\n",
    "\n",
    "print(\"Average difference of {0:6f} with std. dev. of {1:6f}.\"\n",
    "      .format(score_difference.mean(), score_difference.std()))\n",
    "\n",
    "# Plot scores on each trial for nested and non-nested CV\n",
    "plt.figure()\n",
    "plt.subplot(211)\n",
    "non_nested_scores_line, = plt.plot(non_nested_scores, color='r')\n",
    "nested_line, = plt.plot(nested_scores, color='b')\n",
    "plt.ylabel(\"score\", fontsize=\"14\")\n",
    "plt.legend([non_nested_scores_line, nested_line],\n",
    "           [\"Non-Nested CV\", \"Nested CV\"],\n",
    "           bbox_to_anchor=(0, .4, .5, 0))\n",
    "plt.title(\"Non-Nested and Nested Cross Validation on Iris Dataset\",\n",
    "          x=.5, y=1.1, fontsize=\"15\")\n",
    "\n",
    "# Plot bar chart of the difference.\n",
    "plt.subplot(212)\n",
    "difference_plot = plt.bar(range(NUM_TRIALS), score_difference)\n",
    "plt.xlabel(\"Individual Trial #\")\n",
    "plt.legend([difference_plot],\n",
    "           [\"Non-Nested CV - Nested CV Score\"],\n",
    "           bbox_to_anchor=(0, 1, .8, 0))\n",
    "plt.ylabel(\"score difference\", fontsize=\"14\")\n",
    "\n",
    "plt.show()"
   ]
  }
 ],
 "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
}