def test_repe_test(times, features, classes):
every_f1_test = {}
array_means_test = np.array([])
array_vars_test = np.array([])
iii = 0
while iii < times:
x_train, y_train, x_cv, y_cv, x_test, y_test = TrainCvTest. \
separateDatabases(features, classes)
clf = svm.trainSVM(x_train, y_train, call=True)
F1Test = FScore.F1Score(x_test, y_test, clf)
mTest = np.array([])
for key in F1Test.keys():
if key not in every_f1_test:
every_f1_test[key] = F1Test[key]
else:
every_f1_test[key] = np.append(every_f1_test[key], F1Test[key])
mTest = np.append(mTest, F1Test[key])
array_means_test = np.append(array_means_test, mTest.mean())
array_vars_test = np.append(array_vars_test, mTest.var())
iii += 1
return every_f1_test, array_means_test, array_vars_test
def classify_new_sounds(folder_of_test_sounds, target_class):
new_descriptors = {target_class: load_description(folder_of_test_sounds)}
new_normalized_features, new_y_class, new_features_names = preprocessDescriptors(new_descriptors)
new_y_class = np.array([target_class]*len(new_y_class))
descriptors = loadDescriptors(maximum='Inf', reverbs=True)
normalized_features, yClass, features_names = preprocessDescriptors(descriptors)
clf = trainSVM(normalized_features, yClass, call=True)
F1 = F1Score(new_normalized_features, new_y_class, clf)
return F1
def classify_new_sounds(folder_of_test_sounds, target_class):
new_descriptors = {target_class: load_description(folder_of_test_sounds)}
new_normalized_features, new_y_class, new_features_names = preprocessDescriptors(
new_descriptors)
new_y_class = np.array([target_class] * len(new_y_class))
descriptors = loadDescriptors(maximum='Inf', reverbs=True)
normalized_features, yClass, features_names = preprocessDescriptors(
descriptors)
clf = trainSVM(normalized_features, yClass, call=True)
F1 = F1Score(new_normalized_features, new_y_class, clf)
return F1
def recall(XData, YData, clf=1):
if clf == 1:
clf = svm.trainSVM(XData, YData, True)
predictions = []
for iii in np.arange(YData.size):
predictions = np.append(predictions, clf.predict(XData[iii, :])[0])
tp = true_pos(predictions, YData)
fn = false_neg(predictions, YData)
recall = {}
for key in tp.keys():
recall[key] = tp[key] / float(tp[key] + fn[key])
return recall
def test_repe_train_and_cv(times, features, classes):
every_f1_train = {}
every_f1_cv = {}
array_means_train = np.array([])
array_vars_train = np.array([])
array_means_cv = np.array([])
array_vars_cv = np.array([])
iii = 0
while iii < times:
x_train, y_train, x_cv, y_cv, x_test, y_test = TrainCvTest. \
separateDatabases(features, classes)
clf = svm.trainSVM(x_train, y_train, call=True)
F1Train = FScore.F1Score(x_train, y_train, clf)
mTest = np.array([])
for key in F1Train.keys():
if key not in every_f1_cv:
every_f1_train[key] = F1Train[key]
else:
every_f1_train[key] = np.append(every_f1_train[key],
F1Train[key])
mTest = np.append(mTest, F1Train[key])
array_means_train = np.append(array_means_train, mTest.mean())
array_vars_train = np.append(array_vars_train, mTest.var())
F1CV = FScore.F1Score(x_cv, y_cv, clf)
mCV = np.array([])
for key in F1CV.keys():
if key not in every_f1_cv:
every_f1_cv[key] = F1CV[key]
else:
every_f1_cv[key] = np.append(every_f1_cv[key], F1CV[key])
mCV = np.append(mCV, F1CV[key])
array_means_cv = np.append(array_means_cv, mCV.mean())
array_vars_cv = np.append(array_vars_cv, mCV.var())
iii += 1
return every_f1_train, every_f1_cv, array_means_train, \
array_vars_train, array_means_cv, array_vars_cv
def precision(XData, YData, clf=1):
if clf == 1:
clf = svm.trainSVM(XData, YData, True)
predictions = []
for iii in np.arange(YData.size):
predictions = np.append(predictions, clf.predict(XData[iii, :])[0])
tp = true_pos(predictions, YData)
fp = false_pos(predictions, YData)
precision_ = {}
for key in tp.keys():
try:
precision_[key] = tp[key] / float(tp[key] + fp[key])
except:
precision_[key] = -1
return precision_