import numpy as np import read_data as rd from sklearn.cross_validation import KFold train_label,picname_3k,labelname_3k = rd.read_train_3k(rd.train_3k) test_name = rd.read_test_3k(rd.test_3k) attr_name = rd.read_attributes_list(rd.attr_list) train_data = rd.read_attributes(rd.attr_train) # test_data = rd.read_attributes(rd.attr_test) train_data_alex = rd.read_npy(rd.alexnet_train) test_data_alex = rd.read_npy(rd.alexnet_test) train_data_siftbow = rd.read_npy(rd.siftbow_train) # test_data_siftbow = rd.read_npy(rd.siftbow_test) def cross_validate(train_data,n_folds = 5): kf = KFold(len(train_data), n_folds = n_folds) ret = 0.0 for train_index, test_index in kf: X_train, X_test = train_data[train_index], train_data[test_index] y_train, y_test = train_label[train_index], train_label[test_index] temp = svm(X_train,y_train,X_test,y_test) print temp ret += temp return ret/n_folds def svm(X_train,y_train,X_test,y_test): # clf = SVC(C=1.0,kernel='sigmoid') clf = LinearSVC(dual=False) score = clf.fit(X_train,y_train).score(X_test,y_test)
def mode(data):
result = np.zeros((1000, 200))
for i in range(0, len(data)):
result[i][data[i]] = 1
return result
if __name__ == "__main__":
piclabel_3k, picname_3k, labels_3k = read_data.read_train_3k(read_data.train_3k)
attributes_train = read_data.read_attributes(read_data.attr_train)
attributes_test = read_data.read_attributes(read_data.attr_test)
alexnet_train = read_data.read_npy(read_data.alexnet_train)
alexnet_test = read_data.read_npy(read_data.alexnet_test)
alexmin = np.amin(alexnet_train)
alexnet_train_ = alexnet_train - alexmin;
alexnet_test_ = alexnet_test - alexmin
prediction1 = gaussianNB(attributes_train, piclabel_3k, attributes_test)
prediction2 = multinomialNB(attributes_train, piclabel_3k, attributes_test)
prediction3 = bernoulliNB(attributes_train, piclabel_3k, attributes_test)
prediction4 = gaussianNB(alexnet_train, piclabel_3k, alexnet_test)
prediction5 = multinomialNB(alexnet_train_, piclabel_3k, alexnet_test_)
prediction6 = bernoulliNB(alexnet_train, piclabel_3k, alexnet_test)
prediction = mode(prediction1) + mode(prediction2) + mode(prediction3) + mode(prediction4) + mode(prediction5) + mode(prediction6)
prediction = np.argmax(prediction, axis = 1)
test_imgname = read_data.read_test_3k(read_data.test_3k)
csvfile = file('combine_naive_noprob.csv', 'wb')
writer = csv.writer(csvfile)
writer.writerow(["ID", "Category"])
for i in range(0, len(prediction)):
writer.writerow([test_imgname[i],labels_3k[prediction[i]]])
csvfile.close()
import numpy as np
import read_data as rd
from sklearn.cross_validation import KFold
train_label, picname_3k, labelname_3k = rd.read_train_3k(rd.train_3k)
test_name = rd.read_test_3k(rd.test_3k)
attr_name = rd.read_attributes_list(rd.attr_list)
train_data = rd.read_attributes(rd.attr_train)
# test_data = rd.read_attributes(rd.attr_test)
train_data_alex = rd.read_npy(rd.alexnet_train)
test_data_alex = rd.read_npy(rd.alexnet_test)
train_data_siftbow = rd.read_npy(rd.siftbow_train)
# test_data_siftbow = rd.read_npy(rd.siftbow_test)
def cross_validate(train_data, n_folds=5):
kf = KFold(len(train_data), n_folds=n_folds)
ret = 0.0
for train_index, test_index in kf:
X_train, X_test = train_data[train_index], train_data[test_index]
y_train, y_test = train_label[train_index], train_label[test_index]
temp = svm(X_train, y_train, X_test, y_test)
print temp
ret += temp
return ret / n_folds
def svm(X_train, y_train, X_test, y_test):
# clf = SVC(C=1.0,kernel='sigmoid')
clf = LinearSVC(dual=False)
score = clf.fit(X_train, y_train).score(X_test, y_test)