ans2_pros.append(result[i + 1][0])
i += 2
if sorted(ans1_pros, reverse=True)[0] > sorted(ans2_pros,
reverse=True)[0]:
predicty[key][0] = 1
predicty[key + 1][0] = 0
else:
predicty[key][0] = 0
predicty[key + 1][0] = 1
print("update")
return predicty
# train
model = svm.SVC(gamma=10, probability=True)
svm_train(data.trainset, model, 1470)
y, predicty = svm_test(data.testset, model)
eval1 = Evaluation()
eval1.accuracy(y, predicty, data)
with open('result_svm.txt', 'w') as f:
for index, maxd in enumerate(eval1.wrong):
f.write("Case #{}: {} ".format(index + 1, maxd) + '\n')
# predicty=[[0.1], [0.2], [0.1], [0.2], [0.1], [0.2]]
predicty = use_csk(data.testset, predicty, model)
# Evaluation
eval = Evaluation()
eval.accuracy(y, predicty, data)
final = time.time()
print("time:", final - begin)
if __name__ == '__main__':
# datalen = [200, 400, 600, 800, 1000, 1200, 1400, 1600, 1800, 2000,
# 2200, 2400, 2600, 2800, 3000]
# data = Dataset()
# results = []
# for i in datalen:
# model = svm.SVC(gamma=10, probability=True)
# train(data.trainset, model, i)
# y, predicty = test(data.testset, model)
# eval = Evaluation()
# results.append(eval.accuracy(y, predicty, data))
#
# plt.xlabel("Train data text amount")
# plt.ylabel("accuracy")
# plt.plot(datalen, results)
# plt.show()
begin = time.time()
data = Dataset()
model = svm.SVC(gamma=10, probability=True)
train(data.trainset, model, 1470)
y, predicty = test(data.testset, model)
eval = Evaluation()
eval.accuracy(y, predicty, data)
with open('result_svm.txt', 'w') as f:
for index, maxd in enumerate(eval.wrong):
f.write("Case #{}: {} ".format(index + 1, maxd) + '\n')
# final = time.time()
# print("time", final - begin)
def run(self):
results = self.fit_to_validation()
evaluation = Evaluation(results, verbose=False)
evaluation.run()
def main():
# Directory Setting
train_dir = "../data/binary_train.csv"
test_dir = "../data/binary_test.csv"
model_dir = "./model_save"
# HyperParameter
max_len = 50
epoch = 2
batch = 512
hidden_units = 256
# Flow
print("0. Setting Environment")
set_env()
print("1. load data")
train_x, train_y, test_x, test_y, val_x, val_y = load_data(
train_dir, test_dir)
print("2. pre processing")
train_x, val_x, test_x = train_x.tolist(), val_x.tolist(), test_x.tolist()
train_x = [' '.join(t.split()[0:max_len]) for t in train_x]
train_x = np.array(train_x, dtype=object)[:, np.newaxis]
val_x = [' '.join(t.split()[0:max_len]) for t in val_x]
val_x = np.array(val_x, dtype=object)[:, np.newaxis]
test_x = [' '.join(t.split()[0:max_len]) for t in test_x]
test_x = np.array(test_x, dtype=object)[:, np.newaxis]
tokenizer = create_tokenizer_from_hub_module()
train_examples = convert_text_to_examples(train_x, train_y)
val_examples = convert_text_to_examples(val_x, val_y)
test_examples = convert_text_to_examples(test_x, test_y)
train_input_ids, train_input_masks, train_segment_ids, train_labels = convert_examples_to_features(
tokenizer, train_examples, max_len)
val_input_ids, val_input_masks, val_segment_ids, val_labels = convert_examples_to_features(
tokenizer, val_examples, max_len)
test_input_ids, test_input_masks, test_segment_ids, test_labels = convert_examples_to_features(
tokenizer, test_examples, max_len)
print("3. build model")
model = BERT(max_len, data_type="binary")
model.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
initialize_vars(sess)
cp_callback = tf.keras.callbacks.ModelCheckpoint(
filepath=model_dir + "/model-weights.{epoch:02d}-{val_acc:.6f}.hdf5",
monitor='val_acc',
save_best_only=True,
verbose=1)
model.fit(
[train_input_ids, train_input_masks, train_segment_ids],
train_labels,
validation_data=([val_input_ids, val_input_masks,
val_segment_ids], val_labels),
epochs=epoch,
batch_size=batch,
callbacks=[cp_callback])
print("4. evaluation")
evaluation = Evaluation(
model, [test_input_ids, test_input_masks, test_segment_ids], test_y)
accuracy, cf_matrix, report = evaluation.eval_classification_bert(
data_type="binary")
print("## Classification Report \n", report)
print("## Confusion Matrix \n", cf_matrix)
print("## Accuracy \n", accuracy)