<br />
<b>Warning</b>:  Declaration of action_plugin_tablewidth::register(&$controller) should be compatible with DokuWiki_Action_Plugin::register(Doku_Event_Handler $controller) in <b>/s/bach/b/class/cs545/public_html/fall15/lib/plugins/tablewidth/action.php</b> on line <b>93</b><br />
===== model selection and cross validation in scikit-learn =====

First let's import some modules and read in some data:

<code python>

In [1]: import numpy as np

In [2]: from sklearn import cross_validation

In [3]: from sklearn import svm

In [4]: from sklearn import metrics

In [5]: data=np.genfromtxt("../data/heart_scale.data", delimiter=",")

In [6]: X=data[:,1:]

In [7]: y=data[:,0]

</code>

The simplest form of model evaluation uses a validation/test set:

<code python>
In [9]: X_train, X_test, y_train, y_test = cross_validation.train_test_split(X, y, test_size=0.4, random_state=0)

In [10]: classifier = svm.SVC(kernel='linear', C=1).fit(X_train, y_train)

In [11]: classifier.score(X_test, y_test)
Out[11]: 0.7592592592592593


</code>

Next, let'd perform cross-validation:

<code python>

In [18]: cross_validation.cross_val_score(classifier, X, y, cv=5, scoring='accuracy')
Out[18]: array([ 0.7962963 ,  0.83333333,  0.88888889,  0.83333333,  0.83333333])

In [19]: 

In [19]: # you can obtain accuracy for other metrics, such as area under the roc curve:

In [20]: cross_validation.cross_val_score(classifier, X, y, cv=5, scoring='roc_auc')
Out[20]: array([ 0.89166667,  0.89166667,  0.95833333,  0.87638889,  0.91388889])

In [21]: 

In [21]: # you can also obtain the predictions by cross-validation and then compute the accuracy:

In [22]: y_predict = cross_validation.cross_val_predict(classifier, X, y, cv=5)

In [23]: metrics.accuracy_score(y, y_predict)
Out[23]: 0.83703703703703702

</code>

H ere's an alternative way of doing cross-validation.

<code python>
In [25]: # first divide the data into folds:

In [26]: cv = cross_validation.StratifiedKFold(y, 5)

In [27]: # now use these folds:

In [28]: print cross_validation.cross_val_score(classifier, X, y, cv=cv, scoring='roc_auc')
[ 0.89166667  0.89166667  0.95833333  0.87638889  0.91388889]

In [29]: # you can see how examples were divided into folds by looking at the test_folds attribute:

In [30]: print cv.test_folds
[0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2
 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
 2 2 2 2 2 2 2 2 2 2 2 2 3 3 2 3 2 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3
 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 4 4 4 4 4 4
 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4
 4 4 4 4 4 4 4 4 4 4 4]

In [31]: # hmm... perhaps we should shuffle things a bit...

In [32]: cv = cross_validation.StratifiedKFold(y, 5, shuffle=True)

In [33]: print cv.test_folds
[0 1 1 2 0 1 4 3 4 3 2 0 2 3 2 3 2 0 4 1 1 3 4 1 1 4 1 4 4 2 2 3 0 2 3 1 4
 0 3 2 0 2 0 1 3 2 0 0 2 3 0 4 2 0 4 3 4 1 1 0 3 2 4 3 2 3 1 1 1 1 4 3 1 1
 4 2 2 3 3 1 4 2 1 0 2 1 0 2 4 1 0 3 2 3 1 2 2 1 1 0 4 1 3 0 1 1 3 3 0 3 3
 4 2 0 2 0 2 4 0 1 0 4 4 1 1 0 4 0 1 4 4 3 1 3 3 2 4 3 4 2 4 3 4 1 4 2 0 3
 3 3 3 0 0 0 4 3 4 2 3 0 1 1 0 0 4 0 4 1 4 0 0 0 0 3 3 0 4 4 2 0 3 3 0 1 2
 2 2 3 2 1 3 4 4 4 1 1 4 2 1 0 3 1 2 0 0 0 0 2 3 4 3 2 0 0 4 1 3 2 2 0 1 2
 4 2 4 0 2 1 1 0 4 4 1 4 4 3 4 2 3 3 1 4 2 1 4 1 3 2 1 3 2 1 3 1 3 0 2 2 0
 4 4 2 2 4 3 3 0 2 0 2]

In [34]: # if you run division into folds multiple times you will get a different answer:

In [35]: cv = cross_validation.StratifiedKFold(y, 5, shuffle=True)

In [36]: print cv.test_folds
[3 0 2 2 0 2 2 4 1 4 0 2 3 4 2 0 4 0 3 3 4 0 2 0 4 4 0 1 4 4 3 4 1 2 3 3 1
 2 1 4 4 4 0 0 4 2 0 0 2 0 1 3 1 0 3 4 0 3 0 4 1 1 2 4 2 0 2 3 1 0 3 0 1 2
 3 2 4 0 0 0 1 4 3 2 2 4 3 1 3 2 0 2 0 0 3 2 1 2 4 4 0 0 4 2 1 4 3 0 4 3 4
 1 4 0 0 4 2 1 4 4 3 4 1 1 3 0 2 2 3 1 2 3 1 0 4 1 4 1 3 1 3 3 4 4 1 0 0 0
 0 4 3 1 2 2 3 0 3 2 4 3 2 2 3 0 3 1 0 4 2 3 0 2 4 3 0 4 3 4 3 3 0 3 1 2 2
 1 3 4 1 0 4 3 4 0 0 0 3 2 2 1 3 4 4 2 3 4 3 2 1 3 0 4 0 1 3 1 2 2 2 2 0 3
 1 1 1 2 0 1 4 1 1 1 2 2 1 2 3 3 1 4 4 3 4 2 0 2 2 1 1 1 2 0 3 0 2 1 1 3 1
 3 1 0 1 3 4 4 2 1 1 1]

In [37]: # if you want to consistently get the same division into folds:

In [38]: cv = cross_validation.StratifiedKFold(y, 5, shuffle=True, random_state=0)

In [39]: # this sets the seed for the random number generator.

</code>

Let's do grid search for the optimal set of parameters:

<code python>
In [40]: from sklearn.grid_search import GridSearchCV

In [41]: Cs = np.logspace(-2, 3, 6)

In [42]: classifier = GridSearchCV(estimator=svm.LinearSVC(), param_grid=dict(C=Cs) )

In [43]: classifier.fit(X, y)
Out[43]: 
GridSearchCV(cv=None, error_score='raise',
       estimator=LinearSVC(C=1.0, class_weight=None, dual=True, fit_intercept=True,
     intercept_scaling=1, loss='squared_hinge', max_iter=1000,
     multi_class='ovr', penalty='l2', random_state=None, tol=0.0001,
     verbose=0),
       fit_params={}, iid=True, loss_func=None, n_jobs=1,
       param_grid={'C': array([  1.00000e-02,   1.00000e-01,   1.00000e+00,   1.00000e+01,
         1.00000e+02,   1.00000e+03])},
       pre_dispatch='2*n_jobs', refit=True, score_func=None, scoring=None,
       verbose=0)

In [44]: 

In [44]: # print the best accuracy, classifier and parameters:

In [45]: print classifier.best_score_
0.844444444444

In [46]: print classifier.best_estimator_
LinearSVC(C=1.0, class_weight=None, dual=True, fit_intercept=True,
     intercept_scaling=1, loss='squared_hinge', max_iter=1000,
     multi_class='ovr', penalty='l2', random_state=None, tol=0.0001,
     verbose=0)

In [47]: print classifier.best_params_
{'C': 1.0}

n [48]: # performing nested cross validation:

In [49]: print  cross_validation.cross_val_score(classifier, X, y, cv=5)
[ 0.7962963   0.81481481  0.88888889  0.83333333  0.83333333]

In [50]: # if we want to do grid search over multiple parameters:

In [51]: param_grid = [
   ....:   {'C': [1, 10, 100, 1000], 'kernel': ['linear']},
   ....:   {'C': [1, 10, 100, 1000], 'gamma': [0.001, 0.0001], 'kernel': ['rbf']},
   ....:  ]

In [52]: classifier = GridSearchCV(estimator=svm.SVC(), param_grid=param_grid)

In [53]: print cross_validation.cross_val_score(classifier, X, y, cv=5)
[ 0.7962963   0.83333333  0.88888889  0.7962963   0.87037037]

</code>

And to make things easier for you here's the whole thing without the output:

<file python model_selection.py>
import numpy as np
from sklearn import cross_validation
from sklearn import svm
from sklearn import metrics

data=np.genfromtxt("../data/heart_scale.data", delimiter=",")
X=data[:,1:]
y=data[:,0]

# let's train/test an svm on the heart dataset:

X_train, X_test, y_train, y_test = cross_validation.train_test_split(X, y, test_size=0.4, random_state=0)
classifier = svm.SVC(kernel='linear', C=1).fit(X_train, y_train)
print classifier.score(X_test, y_test)

# now let's use cross-validation instead:
print cross_validation.cross_val_score(classifier, X, y, cv=5, scoring='accuracy')

# you can obtain accuracy for other metrics, such as area under the roc curve:
print cross_validation.cross_val_score(classifier, X, y, cv=5, scoring='roc_auc')

# you can also obtain the predictions by cross-validation and then compute the accuracy:
y_predict = cross_validation.cross_val_predict(classifier, X, y, cv=5)
metrics.accuracy_score(y, y_predict)

# here's an alternative way of doing cross-validation.
# first divide the data into folds:
cv = cross_validation.StratifiedKFold(y, 5)
# now use these folds:
print cross_validation.cross_val_score(classifier, X, y, cv=cv, scoring='roc_auc')

# you can see how examples were divided into folds by looking at the test_folds attribute:
print cv.test_folds

# hmm... perhaps we should shuffle things a bit...

cv = cross_validation.StratifiedKFold(y, 5, shuffle=True)
print cv.test_folds

# if you run division into folds multiple times you will get a different answer:
cv = cross_validation.StratifiedKFold(y, 5, shuffle=True)
print cv.test_folds

# if you want to consistently get the same division into folds:
cv = cross_validation.StratifiedKFold(y, 5, shuffle=True, random_state=0)
# this sets the seed for the random number generator.


# grid search

# let's perform model selection using grid search 

from sklearn.grid_search import GridSearchCV
Cs = np.logspace(-2, 3, 6)
classifier = GridSearchCV(estimator=svm.LinearSVC(), param_grid=dict(C=Cs) )
classifier.fit(X, y)

# print the best accuracy, classifier and parameters:
print classifier.best_score_
print classifier.best_estimator_
print classifier.best_params_

# performing nested cross validation:
print  cross_validation.cross_val_score(classifier, X, y, cv=5)

# if we want to do grid search over multiple parameters:
param_grid = [
  {'C': [1, 10, 100, 1000], 'kernel': ['linear']},
  {'C': [1, 10, 100, 1000], 'gamma': [0.001, 0.0001], 'kernel': ['rbf']},
 ]
classifier = GridSearchCV(estimator=svm.SVC(), param_grid=param_grid)
print cross_validation.cross_val_score(classifier, X, y, cv=5)

</file>