y3 = predict2both(y)
print('%s accu:' % i)
# 只取最高置信度的准确率
s1 = [(labels_test[i] == y1[i]).min() for i in range(len(y1))]
print(sum(s1) / len(s1))
# 只取置信度大于0.5的准确率
s2 = [(labels_test[i] == y2[i]).min() for i in range(len(y1))]
print(sum(s2) / len(s2))
# 结合前两个
s3 = [(labels_test[i] == y3[i]).min() for i in range(len(y1))]
print(sum(s3) / len(s3))
print('%s f1:' % i)
# 只取最高置信度的准确率
s4 = f1_avg(y_pred=y1, y_true=labels_test)
print(s4)
# 只取置信度大于0.5的准确率
s5 = f1_avg(y_pred=y2, y_true=labels_test)
print(s5)
# 结合前两个
s6 = f1_avg(y_pred=y3, y_true=labels_test)
print(s6)
score_list1.append(
[i, sum(s1) / len(s1),
sum(s2) / len(s2),
sum(s3) / len(s3)])
score_list2.append([i, s4, s5, s6])
print(pd.DataFrame(score_list1))
print(pd.DataFrame(score_list2))
optimizer = Adam(lr=0.001)
model.compile(optimizer=optimizer, loss='categorical_crossentropy', metrics=['accuracy'])
#####################################################################################################################
n_start = 1
n_end = 51
score_list = []
for i in range(n_start, n_end):
model.fit(x=data_train_x, y=data_train_y, batch_size=batch_size, epochs=1, verbose=2)
# model.save('./models/CNN_simple_3_filters_%d_bs_%d_epochs_%d.h5' % (filters, batch_size, i))
y1 = model.predict(data_valid_x)
y1_tag = p2tag(y1)
valid_tag = p2tag(data_valid_y)
accu = accu_score(y1_tag, valid_tag)
f1 = f1_avg(y1_tag, valid_tag)
score_list.append([i, accu, f1])
metrics = pd.DataFrame(score_list, columns=['batch_size', 'accu', 'f1'])
print(metrics)
metrics.to_csv('./metrics/CNN_simple_3_256_512_1024_filters_%d_bs_%d.csv' % (filters, batch_size),
index=False)
y2 = model.predict(test_x)
y2_tag = p2tag(y2)
result = pd.DataFrame({'ImageId': range(1, 1 + len(y2_tag)), 'Label': y2_tag},
columns=['ImageId', 'Label'])
result.to_csv('./result/result_CNN_simple_3_256_512_1024_filters_%d_bs_%d_epochs_%d.csv' % (filters, batch_size, i),
index=False)
print('end', time.strftime('%Y-%m-%d %H:%M:%S', time.localtime()))