# try using different optimizers and different optimizer configs
model.compile(OPTIMIZER, loss, metrics=['accuracy'])
return model
import encap_sst_bert_zy as models
#embedding_matrix = np.load(('en_aux_files/embeddings_glove_%d.npy' % (VOCAB_SIZE)))
#embedding_matrix=embedding_matrix.T
#model = bd_lstm(embedding_matrix)
#model_path = 'bdlstm_models'
#model.load_weights(model_path)
#test_y2 = np.array([[0, 1] if t == 1 else [1, 0] for t in test_y])
#all_scores_origin = model.evaluate(test_x, test_y2)
#print('all origin test_loss: %f, accuracy: %f' % (all_scores_origin[0], all_scores_origin[1]))
model = models.Model(dataset).cuda()
ga_atttack = PSOAttack(model,word_candidate, dataset,
max_iters=20,
pop_size=pop_size)
SAMPLE_SIZE = len(dataset.test_y)
TEST_SIZE = 1000
test_idx = np.random.choice(len(dataset.test_y), SAMPLE_SIZE, replace=False)
test_len = []
for i in range(SAMPLE_SIZE):
test_len.append(np.sum(np.sign(test_x[test_idx[i]])))
print('Shortest sentence in our test set is %d words' %np.min(test_len))
test_list = []
orig_list = []
orig_label_list = []
adv_list = []
dist_list = []
adv_orig=[]
with open('./all_seqs.pkl', 'rb') as fh:
train, valid, test = pickle.load(fh)
with open('pos_tags_test.pkl', 'rb') as fp:
test_pos_tags = pickle.load(fp)
test_s1 = [t[1:-1] for t in test['s1']]
test_s2 = [t[1:-1] for t in test['s2']]
model = Model()
model.evaluate(test['s1'], test['s2'], test['label'])
np.random.seed(3333)
vocab = {w: i for (w, i) in tokenizer.word_index.items()}
inv_vocab = {i: w for (w, i) in vocab.items()}
#test_accuracy = model.evaluate([test[0], test[1]], test[2])[1]
#print('\nTest accuracy = ', test_accuracy)
adversary = PSOAttack(model, word_candidate, pop_size=60, max_iters=20)
print('the length of test cases is:', len(test_s1))
TEST_SIZE = 5000
test_idxs = np.random.choice(len(test_s1), size=TEST_SIZE, replace=False)
test_list = []
input_list = []
output_list = []
dist_list = []
test_times = []
success = []
change_list = []
target_list = []
true_label_list = []
success_count = 0
i = 0
while len(test_list) < 1000: