Tuesday, May 31, 2016

Python ML 7 - Combining Different Models for Ensemble Learning

-- Learning with ensembles

from scipy.misc import comb
import math
def ensemble_error(n_classifier, error):
     k_start = math.ceil(n_classifier / 2.0)
     probs = [comb(n_classifier, k) *
              error**k *
              (1-error)**(n_classifier - k)
              for k in range(k_start, n_classifier + 1)]
     return sum(probs)
ensemble_error(n_classifier=11, error=0.25)

import numpy as np
error_range = np.arange(0.0, 1.01, 0.01)
ens_errors = [ensemble_error(n_classifier=11, error=error)
               for error in error_range]
import matplotlib.pyplot as plt
plt.plot(error_range, ens_errors,
          label='Ensemble error',
          linewidth=2)
plt.plot(error_range, error_range,
          linestyle='--', label='Base error',
          linewidth=2)
plt.xlabel('Base error')
plt.ylabel('Base/Ensemble error')
plt.legend(loc='upper left')
plt.grid()
plt.show()

-- Implementing a simple majority vote classifier

import numpy as np
np.argmax(np.bincount([0, 0, 1],
           weights=[0.2, 0.2, 0.6]))

ex = np.array([[0.9, 0.1],
                [0.8, 0.2],
                [0.4, 0.6]])
p = np.average(ex, axis=0, weights=[0.2, 0.2, 0.6])
p
np.argmax(p)

ex = np.array([[0.9, 0.1],
                [0.8, 0.2],
                [0.4, 0.6]])
p = np.average(ex, axis=0, weights=[0.2, 0.2, 0.6])
p
np.argmax(p)

from sklearn.base import BaseEstimator
from sklearn.base import ClassifierMixin
from sklearn.preprocessing import LabelEncoder
from sklearn.externals import six
from sklearn.base import clone
from sklearn.pipeline import _name_estimators
import numpy as np
import operator


class MajorityVoteClassifier(BaseEstimator,
                             ClassifierMixin):
    """ A majority vote ensemble classifier

    Parameters
    ----------
    classifiers : array-like, shape = [n_classifiers]
      Different classifiers for the ensemble

    vote : str, {'classlabel', 'probability'}
      Default: 'classlabel'
      If 'classlabel' the prediction is based on
      the argmax of class labels. Else if
      'probability', the argmax of the sum of
      probabilities is used to predict the class label
      (recommended for calibrated classifiers).

    weights : array-like, shape = [n_classifiers]
      Optional, default: None
      If a list of `int` or `float` values are
      provided, the classifiers are weighted by
      importance; Uses uniform weights if `weights=None`.

    """
    def __init__(self, classifiers,
                 vote='classlabel', weights=None):

        self.classifiers = classifiers
        self.named_classifiers = {key: value for
                                  key, value in
                                  _name_estimators(classifiers)}
        self.vote = vote
        self.weights = weights

    def fit(self, X, y):
        """ Fit classifiers.

        Parameters
        ----------
        X : {array-like, sparse matrix},
            shape = [n_samples, n_features]
            Matrix of training samples.

        y : array-like, shape = [n_samples]
            Vector of target class labels.

        Returns
        -------
        self : object

        """
        # Use LabelEncoder to ensure class labels start
        # with 0, which is important for np.argmax
        # call in self.predict
        self.lablenc_ = LabelEncoder()
        self.lablenc_.fit(y)
        self.classes_ = self.lablenc_.classes_
        self.classifiers_ = []
        for clf in self.classifiers:
            fitted_clf = clone(clf).fit(X,
                              self.lablenc_.transform(y))
            self.classifiers_.append(fitted_clf)
        return self


def predict(self, X):
        """ Predict class labels for X.

        Parameters
        ----------
        X : {array-like, sparse matrix},
            Shape = [n_samples, n_features]
            Matrix of training samples.

        Returns
        ----------
        maj_vote : array-like, shape = [n_samples]
            Predicted class labels.

        """
        if self.vote == 'probability':
            maj_vote = np.argmax(self.predict_proba(X),
                                 axis=1)
        else:  # 'classlabel' vote

            #  Collect results from clf.predict calls
            predictions = np.asarray([clf.predict(X)
                                      for clf in
                                      self.classifiers_]).T

            maj_vote = np.apply_along_axis(
                           lambda x:
                           np.argmax(np.bincount(x,                                          
                                        weights=self.weights)),
                           axis=1,
                           arr=predictions)
        maj_vote = self.lablenc_.inverse_transform(maj_vote)
        return maj_vote

    def predict_proba(self, X):
        """ Predict class probabilities for X.

        Parameters
        ----------
        X : {array-like, sparse matrix},
            shape = [n_samples, n_features]
            Training vectors, where n_samples is
            the number of samples and
            n_features is the number of features.

        Returns
        ----------
        avg_proba : array-like,
            shape = [n_samples, n_classes]
            Weighted average probability for
            each class per sample.

        """
        probas = np.asarray([clf.predict_proba(X)
                             for clf in self.classifiers_])
        avg_proba = np.average(probas,
                               axis=0, weights=self.weights)
        return avg_proba

    def get_params(self, deep=True):
        """ Get classifier parameter names for GridSearch"""
        if not deep:
            return super(MajorityVoteClassifier,
                         self).get_params(deep=False)
        else:
            out = self.named_classifiers.copy()
            for name, step in\
                    six.iteritems(self.named_classifiers):
                for key, value in six.iteritems(
                        step.get_params(deep=True)):
                    out['%s__%s' % (name, key)] = value
            return out

Combining different algorithms for classification with majority vote

from sklearn import datasets
from sklearn.cross_validation import train_test_split
from sklearn.preprocessing import StandardScaler
from sklearn.preprocessing import LabelEncoder
iris = datasets.load_iris()
X, y = iris.data[50:, [1, 2]], iris.target[50:]
le = LabelEncoder()
y = le.fit_transform(y)

X_train, X_test, y_train, y_test =\
        train_test_split(X, y,
                         test_size=0.5,
                         random_state=1)

from sklearn.cross_validation import cross_val_score
from sklearn.linear_model import LogisticRegression
from sklearn.tree import DecisionTreeClassifier
from sklearn.neighbors import KNeighborsClassifier
from sklearn.pipeline import Pipeline
import numpy as np
clf1 = LogisticRegression(penalty='l2',
                           C=0.001,
                           random_state=0)
clf2 = DecisionTreeClassifier(max_depth=1,
                               criterion='entropy',
                               random_state=0)
clf3 = KNeighborsClassifier(n_neighbors=1,
                             p=2,
                             metric='minkowski')
pipe1 = Pipeline([['sc', StandardScaler()],
                   ['clf', clf1]])
pipe3 = Pipeline([['sc', StandardScaler()],
                   ['clf', clf3]])
clf_labels = ['Logistic Regression', 'Decision Tree', 'KNN']
print('10-fold cross validation:\n')
for clf, label in zip([pipe1, clf2, pipe3], clf_labels):
     scores = cross_val_score(estimator=clf,
                              X=X_train,
                              y=y_train,
                              cv=10,
                              scoring='roc_auc')
     print("ROC AUC: %0.2f (+/- %0.2f) [%s]"
                % (scores.mean(), scores.std(), label))


mv_clf = MajorityVoteClassifier(
                 classifiers=[pipe1, clf2, pipe3])
clf_labels += ['Majority Voting']
all_clf = [pipe1, clf2, pipe3, mv_clf]
for clf, label in zip(all_clf, clf_labels):
     scores = cross_val_score(estimator=clf,
                              X=X_train,
                              y=y_train,
                              cv=10,
                              scoring='roc_auc')
     print("Accuracy: %0.2f (+/- %0.2f) [%s]"
                % (scores.mean(), scores.std(), label))


-- Evaluating and tuning the ensemble classifier

from sklearn.metrics import roc_curve
from sklearn.metrics import auc
colors = ['black', 'orange', 'blue', 'green']
linestyles = [':', '--', '-.', '-']
for clf, label, clr, ls \
         in zip(all_clf, clf_labels, colors, linestyles):
     # assuming the label of the positive class is 1
     y_pred = clf.fit(X_train,
                      y_train).predict_proba(X_test)[:, 1]
     fpr, tpr, thresholds = roc_curve(y_true=y_test,
                                      y_score=y_pred)
     roc_auc = auc(x=fpr, y=tpr)
     plt.plot(fpr, tpr,
              color=clr,
              linestyle=ls,
              label='%s (auc = %0.2f)' % (label, roc_auc))
plt.legend(loc='lower right')
plt.plot([0, 1], [0, 1],
          linestyle='--',
          color='gray',
          linewidth=2)
plt.xlim([-0.1, 1.1])
plt.ylim([-0.1, 1.1])
plt.grid()
plt.xlabel('False Positive Rate')
plt.ylabel('True Positive Rate')
plt.show()

sc = StandardScaler()
X_train_std = sc.fit_transform(X_train)
from itertools import product
x_min = X_train_std[:, 0].min() - 1
x_max = X_train_std[:, 0].max() + 1
y_min = X_train_std[:, 1].min() - 1
y_max = X_train_std[:, 1].max() + 1
xx, yy = np.meshgrid(np.arange(x_min, x_max, 0.1),
                      np.arange(y_min, y_max, 0.1))
f, axarr = plt.subplots(nrows=2, ncols=2,
                         sharex='col',
                         sharey='row',
                         figsize=(7, 5))
for idx, clf, tt in zip(product([0, 1], [0, 1]),
                         all_clf, clf_labels):
     clf.fit(X_train_std, y_train)
     Z = clf.predict(np.c_[xx.ravel(), yy.ravel()])
     Z = Z.reshape(xx.shape)
     axarr[idx[0], idx[1]].contourf(xx, yy, Z, alpha=0.3)  
     axarr[idx[0], idx[1]].scatter(X_train_std[y_train==0, 0],
                                   X_train_std[y_train==0, 1],
                                   c='blue',
                                   marker='^',
                                   s=50)  
     axarr[idx[0], idx[1]].scatter(X_train_std[y_train==1, 0],
                                   X_train_std[y_train==1, 1],
                                   c='red',
                                   marker='o',
                                   s=50)
     axarr[idx[0], idx[1]].set_title(tt)
plt.text(-3.5, -4.5,
          s='Sepal width [standardized]',
          ha='center', va='center', fontsize=12)
plt.text(-10.5, 4.5,
          s='Petal length [standardized]',
          ha='center', va='center',
          fontsize=12, rotation=90)
plt.show()

how to access the individual parameters inside a Grid Search Object:
mv_clf.get_params()

from sklearn.grid_search import GridSearchCV
params = {'decisiontreeclassifier__max_depth': [1, 2],
           'pipeline-1__clf__C': [0.001, 0.1, 100.0]}
grid = GridSearchCV(estimator=mv_clf,
                     param_grid=params,
                     cv=10,
                     scoring='roc_auc')
grid.fit(X_train, y_train)

for params, mean_score, scores in grid.grid_scores_:
     print("%0.3f+/-%0.2f %r"
            % (mean_score, scores.std() / 2, params))

print('Best parameters: %s' % grid.best_params_)

print('Accuracy: %.2f' % grid.best_score_)

-- Bagging - building an ensemble of classifier from bootstrap samples

import pandas as pd
df_wine = pd.read_csv('https://archive.ics.uci.edu/ml/machine-learning-databases/wine/wine.data', header=None)
df_wine.columns = ['Class label', 'Alcohol',
                    'Malic acid', 'Ash',
                    'Alcalinity of ash',
                    'Magnesium', 'Total phenols',
                    'Flavanoids', 'Nonflavanoid phenols',
                    'Proanthocyanins',
                    'Color intensity', 'Hue',
                    'OD280/OD315 of diluted wines',
                    'Proline']
df_wine = df_wine[df_wine['Class label'] != 1]
y = df_wine['Class label'].values
X = df_wine[['Alcohol', 'Hue']].values

from sklearn.preprocessing import LabelEncoder
from sklearn.cross_validation import train_test_split
le = LabelEncoder()
y = le.fit_transform(y)
X_train, X_test, y_train, y_test =\
            train_test_split(X, y,
                             test_size=0.40,
                             random_state=1)

from sklearn.ensemble import BaggingClassifier
tree = DecisionTreeClassifier(criterion='entropy',
                               max_depth=None)
bag = BaggingClassifier(base_estimator=tree,
                         n_estimators=500,
                         max_samples=1.0,
                         max_features=1.0,
                         bootstrap=True,
                         bootstrap_features=False,
                         n_jobs=1,
                         random_state=1)

compare the performance of the bagging classifier to the performance of a single unpruned decision tree
from sklearn.metrics import accuracy_score
tree = tree.fit(X_train, y_train)
y_train_pred = tree.predict(X_train)
y_test_pred = tree.predict(X_test)
tree_train = accuracy_score(y_train, y_train_pred)
tree_test = accuracy_score(y_test, y_test_pred)
print('Decision tree train/test accuracies %.3f/%.3f'
        % (tree_train, tree_test))

bag = bag.fit(X_train, y_train)
y_train_pred = bag.predict(X_train)
y_test_pred = bag.predict(X_test)
bag_train = accuracy_score(y_train, y_train_pred)
bag_test = accuracy_score(y_test, y_test_pred)
print('Bagging train/test accuracies %.3f/%.3f'
        % (bag_train, bag_test))

x_min = X_train[:, 0].min() - 1
x_max = X_train[:, 0].max() + 1
y_min = X_train[:, 1].min() - 1
y_max = X_train[:, 1].max() + 1
xx, yy = np.meshgrid(np.arange(x_min, x_max, 0.1),
                      np.arange(y_min, y_max, 0.1))
f, axarr = plt.subplots(nrows=1, ncols=2,
                         sharex='col',
                         sharey='row',
                         figsize=(8, 3))
for idx, clf, tt in zip([0, 1],
                         [tree, bag],
                         ['Decision Tree', 'Bagging']):
     clf.fit(X_train, y_train)
   
     Z = clf.predict(np.c_[xx.ravel(), yy.ravel()])
     Z = Z.reshape(xx.shape)
     axarr[idx].contourf(xx, yy, Z, alpha=0.3)
     axarr[idx].scatter(X_train[y_train==0, 0],
                        X_train[y_train==0, 1],
                        c='blue', marker='^')  
     axarr[idx].scatter(X_train[y_train==1, 0],
                        X_train[y_train==1, 1],
                        c='red', marker='o')  
     axarr[idx].set_title(tt)
axarr[0].set_ylabel(Alcohol', fontsize=12)
plt.text(10.2, -1.2,
          s=Hue',
          ha='center', va='center', fontsize=12)
plt.show()

-- Leveraging weak learners via adaptive boosting

from sklearn.ensemble import AdaBoostClassifier
tree = DecisionTreeClassifier(criterion='entropy',
                               max_depth=1)
ada = AdaBoostClassifier(base_estimator=tree,
                          n_estimators=500,
                          learning_rate=0.1,
                          random_state=0)
tree = tree.fit(X_train, y_train)
y_train_pred = tree.predict(X_train)
y_test_pred = tree.predict(X_test)
tree_train = accuracy_score(y_train, y_train_pred)
tree_test = accuracy_score(y_test, y_test_pred)
print('Decision tree train/test accuracies %.3f/%.3f'
       % (tree_train, tree_test))

ada = ada.fit(X_train, y_train)
y_train_pred = ada.predict(X_train)
y_test_pred = ada.predict(X_test)
ada_train = accuracy_score(y_train, y_train_pred)
ada_test = accuracy_score(y_test, y_test_pred)
print('AdaBoost train/test accuracies %.3f/%.3f'
       % (ada_train, ada_test))


x_min = X_train[:, 0].min() - 1
x_max = X_train[:, 0].max() + 1
y_min = X_train[:, 1].min() - 1
y_max = X_train[:, 1].max() + 1
xx, yy = np.meshgrid(np.arange(x_min, x_max, 0.1),
                      np.arange(y_min, y_max, 0.1))
f, axarr = plt.subplots(1, 2,
                         sharex='col',
                         sharey='row',
                         figsize=(8, 3))
for idx, clf, tt in zip([0, 1],
                         [tree, ada],
                         ['Decision Tree', 'AdaBoost']):
     clf.fit(X_train, y_train)
     Z = clf.predict(np.c_[xx.ravel(), yy.ravel()])
     Z = Z.reshape(xx.shape)
     axarr[idx].contourf(xx, yy, Z, alpha=0.3)
     axarr[idx].scatter(X_train[y_train==0, 0],
                        X_train[y_train==0, 1],
                        c='blue',
                        marker='^')
     axarr[idx].scatter(X_train[y_train==1, 0],
                        X_train[y_train==1, 1],
                        c='red',
                        marker='o')
 axarr[idx].set_title(tt)
 axarr[0].set_ylabel('Alcohol', fontsize=12)
plt.text(10.2, -1.2,
          s=Hue',
          ha='center',
          va='center',
          fontsize=12)  
plt.show()

No comments:

Post a Comment