local_x, global_x, y, tag = cPickle.load(
open(alignedFeaturesFilename, 'rb'))
y = label_binarize(y, classes=range(7))
n_classes = 7
print("Training SVM")
TIMES = 100
l = []
for i in range(TIMES):
print '\rFitting %d/%d ' % (i, TIMES),
sys.stdout.flush()
# resampling
classifier = OneVsRestClassifier(
svm.SVC(kernel=multichannel_wrapper(2, chi_square_kernel),
probability=True))
X_train, X_test, y_train, y_test = train_test_split(
local_x, global_x, y, tag)
y_score = classifier.fit(X_train, y_train).decision_function(X_test)
l.append(
float((y_test.argmax(1) == y_score.argmax(1)).sum()) /
y_score.shape[0] * 100)
print map(lambda x: '%.3f%%' % x, l), '=', np.mean(l)
y_score = classifier.decision_function(x)
print 'Test all = %.3f%%' % (float(
(y.argmax(1) == y_score.argmax(1)).sum()) / y_score.shape[0] * 100)
if True:
y = np.array(y, dtype=int)-1
y = label_binarize(y, classes=range(7))
n_classes = 7
X_train, X_test, y_train, y_test = train_test_split(x, y, tag)
print("Training SVM")
TIMES = 10
l = []
for i in range(TIMES):
print '\rFitting %d/%d ' % (i, TIMES),
sys.stdout.flush()
# resampling
classifier = OneVsRestClassifier(svm.SVC(kernel=multichannel_wrapper(2, chi_square_kernel), probability=True))
X_train, X_test, y_train, y_test = train_test_split(x, y, tag)
y_score = classifier.fit(X_train, y_train).decision_function(X_test)
l.append(float((y_test.argmax(1) == y_score.argmax(1)).sum())/y_score.shape[0]*100)
print map(lambda x: '%.3f%%' % x, l), '=', np.mean(l)
y_score = classifier.decision_function(x)
print 'Test all = %.3f%%' % (float((y.argmax(1) == y_score.argmax(1)).sum())/y_score.shape[0]*100 )
if True:
import matplotlib.pyplot as plt
from sklearn.metrics import precision_recall_curve
from sklearn.metrics import average_precision_score
# Compute Precision-Recall and plot curve
