def multiclass(train_feature, train_label, test_feature, clftype, method_name,
paras):
""" The multi classifier method
clftype: 'multiclass', 'onevsrest', 'onevsone'
method_name: the classifier name
paras: list form of parameters
"""
Classifier, kwargs = get_classifier_by_name(method_name, paras)
print 'Method: ', method_name
from sklearn.cross_validation import KFold
kf = KFold(len(train_label), n_folds, indices=True)
index = 0
avg_f1_score_list = [0] * n_folds
for train_index, test_index in kf:
print 'Prepare cv dataset: %d' % index
model_train_feature = train_feature[train_index, :]
model_test_feature = train_feature[test_index, :]
model_train_label = train_label[train_index]
model_test_label = train_label[test_index]
#print 'Over sampling...'
#model_train_feature, model_train_label = over_sampling(model_train_feature, model_train_label)
#ipdb.set_trace()
print 'SMOTE over sampling...'
model_train_feature, model_train_label = smote_sampling(
model_train_feature, model_train_label)
clf = get_classifier_by_type(clftype, model_train_feature,
model_train_label, Classifier, kwargs)
model_test_pred = clf.predict(model_test_feature)
print 'Model testing acc:'
print classification_report(model_test_label, model_test_pred)
#f1_score_list = f1_score(model_test_label, model_test_pred, average=None)
#avg_f1_score_list[index] = sum(f1_score_list) / len(f1_score_list)
#print 'F1 score:', f1_score_list, 'Avg:', avg_f1_score_list[index]
avg_f1_score_list[index] = f1_score_dict(
predition2dict(model_test_pred), predition2dict(model_test_label))
print 'Avg: ', avg_f1_score_list[index]
index += 1
print 'Method:', method_name
avg_avg_f1_score = sum(avg_f1_score_list) / len(avg_f1_score_list)
print 'Avg avg_f1_score:', avg_avg_f1_score, '\n'
#print 'Oversampling...'
#train_feature, train_label = over_sampling(train_feature, train_label)
print 'SMOTE over sampling...'
train_feature, train_label = smote_sampling(train_feature, train_label)
print 'Train the whole multi-class classifiers...'
clf = get_classifier_by_type(clftype, train_feature, train_label,
Classifier, kwargs)
train_pred = clf.predict(train_feature)
test_pred = clf.predict(test_feature)
print 'Model train acc:'
print classification_report(train_label, train_pred)
#f1_score_list = f1_score(train_label, train_pred, average=None)
#avg_f1_score = sum(f1_score_list) / len(f1_score_list)
#print 'F1 score:', f1_score_list, 'Avg:', avg_f1_score
# training F1 score
print 'Training avg F1 score:', f1_score_dict(predition2dict(train_pred),
predition2dict(train_label))
return method_name, test_pred, avg_avg_f1_score
def main(n_components, n_folds, method_name):
print 'Load dataset...'
import pickle
f = open('task2-dataset/task2-dataset.pickle', 'r')
train_feature, train_label, test_feature = pickle.load(f)
f.close()
train_feature, test_feature = PCA_transform(
train_feature, test_feature, 'task2-dataset/task2-PCA-decomp.mat')
train_feature = train_feature[:, :n_components]
test_feature = test_feature[:, :n_components]
kwargs = {}
#from sklearn.naive_bayes import GaussianNB as Classifier
#method_name = 'Twostep+NB'
#from sklearn.svm import LinearSVC as Classifier
#method_name = 'Twostep+SVC'
#kwargs = {'random_state':0, 'C':10}
from QDF import QDF as Classifier
method_name = 'Twostep+QDF'
#from LDF import LDF as Classifier
#method_name = 'Twostep+LDF'
print 'Method: ', method_name
from sklearn.cross_validation import KFold
kf = KFold(len(train_label), n_folds, indices=True)
index = 0
avg_f1_score_list = [0] * n_folds
for train_index, test_index in kf:
print 'Prepare cv dataset: %d' % index
model_train_feature = train_feature[train_index, :]
model_test_feature = train_feature[test_index, :]
model_train_label = train_label[train_index]
model_test_label = train_label[test_index]
model_train_pred, model_test_pred = twostep(model_train_feature,
model_train_label,
model_test_feature,
Classifier, kwargs)
#print 'Model testing acc:'
#print classification_report(model_test_label, model_test_pred)
#f1_score_list = f1_score(model_test_label, model_test_pred, average=None)
#avg_f1_score_list[index] = sum(f1_score_list) / len(f1_score_list)
#print 'F1 score:', f1_score_list, 'Avg:', avg_f1_score_list[index]
avg_f1_score_list[index] = f1_score_dict(
predition2dict(model_test_pred), predition2dict(model_test_label))
print 'Avg: ', avg_f1_score_list[index]
index += 1
print 'Method:', method_name
avg_avg_f1_score = sum(avg_f1_score_list) / len(avg_f1_score_list)
print 'Avg avg_f1_score:', avg_avg_f1_score, '\n'
print 'Train the whole multi-class classifiers...'
train_pred, test_pred = twostep(train_feature, train_label, test_feature,
Classifier, kwargs)
# training F1 score
print 'Training avg F1 score:', f1_score_dict(predition2dict(train_pred),
predition2dict(train_label))
# save the final prediction
index = 0
f = open('twostep_output.csv', 'w')
for y in test_pred:
f.write('%d,%d\n' % (index + 1, test_pred[index]))
index += 1
f.close()
def main(n_components, n_folds, method_name):
print 'Load dataset...'
import pickle
f = open('task2-dataset/task2-dataset.pickle', 'r')
train_feature, train_label, test_feature = pickle.load(f)
f.close()
train_feature, test_feature = PCA_transform(train_feature, test_feature, 'task2-dataset/task2-PCA-decomp.mat')
train_feature = train_feature[:, :n_components]
test_feature = test_feature[:, :n_components]
kwargs = {}
#from sklearn.naive_bayes import GaussianNB as Classifier
#method_name = 'Twostep+NB'
#from sklearn.svm import LinearSVC as Classifier
#method_name = 'Twostep+SVC'
#kwargs = {'random_state':0, 'C':10}
from QDF import QDF as Classifier
method_name = 'Twostep+QDF'
#from LDF import LDF as Classifier
#method_name = 'Twostep+LDF'
print 'Method: ', method_name
from sklearn.cross_validation import KFold
kf = KFold(len(train_label), n_folds, indices=True)
index = 0
avg_f1_score_list = [0] * n_folds
for train_index, test_index in kf:
print 'Prepare cv dataset: %d' % index
model_train_feature = train_feature[train_index, :]
model_test_feature = train_feature[test_index, :]
model_train_label = train_label[train_index]
model_test_label = train_label[test_index]
model_train_pred, model_test_pred = twostep(model_train_feature, model_train_label, model_test_feature, Classifier, kwargs)
#print 'Model testing acc:'
#print classification_report(model_test_label, model_test_pred)
#f1_score_list = f1_score(model_test_label, model_test_pred, average=None)
#avg_f1_score_list[index] = sum(f1_score_list) / len(f1_score_list)
#print 'F1 score:', f1_score_list, 'Avg:', avg_f1_score_list[index]
avg_f1_score_list[index] = f1_score_dict(predition2dict(model_test_pred), predition2dict(model_test_label))
print 'Avg: ', avg_f1_score_list[index]
index += 1
print 'Method:', method_name
avg_avg_f1_score = sum(avg_f1_score_list) / len(avg_f1_score_list)
print 'Avg avg_f1_score:', avg_avg_f1_score, '\n'
print 'Train the whole multi-class classifiers...'
train_pred, test_pred = twostep(train_feature, train_label, test_feature, Classifier, kwargs)
# training F1 score
print 'Training avg F1 score:', f1_score_dict(predition2dict(train_pred), predition2dict(train_label))
# save the final prediction
index = 0
f = open('twostep_output.csv', 'w')
for y in test_pred:
f.write('%d,%d\n' % (index+1, test_pred[index]))
index += 1
f.close()
def multiclass(train_feature, train_label, test_feature, clftype, method_name, paras):
""" The multi classifier method
clftype: 'multiclass', 'onevsrest', 'onevsone'
method_name: the classifier name
paras: list form of parameters
"""
Classifier, kwargs = get_classifier_by_name(method_name, paras)
print 'Method: ', method_name
from sklearn.cross_validation import KFold
kf = KFold(len(train_label), n_folds, indices=True)
index = 0
avg_f1_score_list = [0] * n_folds
for train_index, test_index in kf:
print 'Prepare cv dataset: %d' % index
model_train_feature = train_feature[train_index, :]
model_test_feature = train_feature[test_index, :]
model_train_label = train_label[train_index]
model_test_label = train_label[test_index]
#print 'Over sampling...'
#model_train_feature, model_train_label = over_sampling(model_train_feature, model_train_label)
#ipdb.set_trace()
print 'SMOTE over sampling...'
model_train_feature, model_train_label = smote_sampling(model_train_feature, model_train_label)
clf = get_classifier_by_type(clftype, model_train_feature, model_train_label, Classifier, kwargs)
model_test_pred = clf.predict(model_test_feature)
print 'Model testing acc:'
print classification_report(model_test_label, model_test_pred)
#f1_score_list = f1_score(model_test_label, model_test_pred, average=None)
#avg_f1_score_list[index] = sum(f1_score_list) / len(f1_score_list)
#print 'F1 score:', f1_score_list, 'Avg:', avg_f1_score_list[index]
avg_f1_score_list[index] = f1_score_dict(predition2dict(model_test_pred), predition2dict(model_test_label))
print 'Avg: ', avg_f1_score_list[index]
index += 1
print 'Method:', method_name
avg_avg_f1_score = sum(avg_f1_score_list) / len(avg_f1_score_list)
print 'Avg avg_f1_score:', avg_avg_f1_score, '\n'
#print 'Oversampling...'
#train_feature, train_label = over_sampling(train_feature, train_label)
print 'SMOTE over sampling...'
train_feature, train_label = smote_sampling(train_feature, train_label)
print 'Train the whole multi-class classifiers...'
clf = get_classifier_by_type(clftype, train_feature, train_label, Classifier, kwargs)
train_pred = clf.predict(train_feature)
test_pred = clf.predict(test_feature)
print 'Model train acc:'
print classification_report(train_label, train_pred)
#f1_score_list = f1_score(train_label, train_pred, average=None)
#avg_f1_score = sum(f1_score_list) / len(f1_score_list)
#print 'F1 score:', f1_score_list, 'Avg:', avg_f1_score
# training F1 score
print 'Training avg F1 score:', f1_score_dict(predition2dict(train_pred), predition2dict(train_label))
return method_name, test_pred, avg_avg_f1_score
