train_y = np.array(dataset.train_y)
test_x = pad_sequences(dataset.test_seqs2, maxlen=max_len, padding='post')
test_y = np.array(dataset.test_y)
model = models.Model(dataset, model_path).cuda()
attacker = LSAttack(model, word_candidate)
all_test_num = len(dataset.test_y)
print(f'Total have {all_test_num} test examples')
wrong_classified_num = 0
total_num = 0
# logger
cur_logger = MyLogger(log_file)
# save results
test_idx_list = []
cur_saver = ResultSaver()
log_file.write('======================START======================\n\n')
for test_idx in range(all_test_num):
total_num += 1
pos_tags = test_pos_tags[test_idx]
x_orig = test_x[test_idx]
orig_label = test_y[test_idx]
orig_preds = model.predict(x_orig[np.newaxis, :])[0]
if np.argmax(orig_preds) != orig_label:
word_candidate,
pop_size=pop_size,
max_iters=max_iter)
print('the length of test cases is:', len(test_s1))
test_idxs = np.arange(len(test_s1))
# np.random.shuffle(test_idxs)
# test_idxs = np.random.choice(len(test_s1), size=len(test_s1), replace=False)
test_list = []
skip_num = 0
total_num = 0
# logger
cur_logger = MyLogger(log_file=log_file)
# saver
cur_saver = ResultSaver()
print('start')
for test_idx in test_idxs:
total_num += 1
s1 = test_s1[test_idx]
s2 = test_s2[test_idx]
pos_tags = test_pos_tags[test_idx]
attack_pred = np.argmax(model.pred([s1], [s2])[0])
true_label = test['label'][test_idx]
x_len = np.sum(np.sign(s2))
if attack_pred != true_label:
print('Wrong classified')