def predict(path):
np.random.seed(42)
tf.set_random_seed(42)
config = Config.get_default_config(args)
model = Model(config)
predictor = InteractivePredictor(config, model, path)
predictor.predict()
model.close_session()
args = parser.parse_args()
replace_tokens = ["@R_%d@"%i for i in range(1, args.num_replace_tokens+1)]
if args.debug:
config = Config.get_debug_config(args)
else:
config = Config.get_default_config(args)
# Composite training loss
lamb = args.lamb
print('Lamb :=' + str(lamb))
model = Model(config, replace_tokens)
print('Created model')
if config.TRAIN_PATH:
model.train(lamb=lamb)
if config.TEST_PATH and not args.data_path:
results, precision, recall, f1 = model.evaluate()
print('Accuracy: ' + str(results))
print('Precision: ' + str(precision) + ', recall: ' + str(recall) + ', F1: ' + str(f1))
if args.predict:
predictor = InteractivePredictor(config, model)
predictor.predict()
if args.adv_eval:
model.adv_eval_batched()
if args.release and args.load_path:
model.evaluate(release=True)
model.close_session()
parser.add_argument('--predict', action='store_true')
parser.add_argument('--debug', action='store_true')
parser.add_argument('--seed', type=int, default=239)
parser.add_argument('--reports', type=str, default=239)
args = parser.parse_args()
np.random.seed(args.seed)
tf.set_random_seed(args.seed)
if args.debug:
config = Config.get_debug_config(args)
else:
config = Config.get_default_config(args)
model = Model(config)
print('Created model')
if config.TRAIN_PATH:
model.train()
if config.TEST_PATH and not args.data_path:
results, precision, recall, f1, rouge = model.evaluate()
print('Accuracy: ' + str(results))
print('Precision: ' + str(precision) + ', recall: ' + str(recall) + ', F1: ' + str(f1))
print('Rouge: ', rouge)
if args.predict:
predictor = InteractivePredictor(config, model, args.reports)
predictor.predict()
if args.release and args.load_path:
model.evaluate(release=True)
model.close_session()
parser.add_argument('--predict', action='store_true')
parser.add_argument('--debug', action='store_true')
args = parser.parse_args()
if args.debug:
config = Config.get_debug_config(args)
else:
config = Config.get_default_config(args)
model = Model(config)
print('Created model')
if config.TRAIN_PATH:
model.train()
if config.TEST_PATH and not args.data_path:
results, precision, recall, f1 = model.evaluate()
print('Accuracy: ' + str(results))
print('Precision: ' + str(precision) + ', recall: ' + str(recall) + ', F1: ' + str(f1))
if args.predict:
print("Under Prediciton process.....")
predictor = InteractivePredictor(config, model)
if extract_AST:
ast_fetaure_list = predictor.get_ast_paths_for_file()
print(f"AST Feature for {len(ast_fetaure_list)} snippets")
print(ast_fetaure_list)
else:
predictor.predict()
if args.release and args.load_path:
model.evaluate(release=True)
model.close_session()
def predict(self):
input_filename = 'Input.java'
MAX_ATTEMPTS = 50
MAX_NODES_TO_OPEN = 10
words_vocab = self.model.get_words_vocab_embed()
# words_vocab = words_vocab / np.linalg.norm(words_vocab, axis=1).reshape((-1, 1))
print('Starting interactive prediction with similar adversarial search...')
while True:
print(
'Modify the file: "%s" and press any key when ready, or "q" / "quit" / "exit" to exit' % input_filename)
user_input = input()
if user_input.lower() in self.exit_keywords:
print('Exiting...')
return
print("select variable to rename:")
var_to_rename = newname_of_var = input()
name_found = False
closed = [var_to_rename]
with open(input_filename, "r") as f:
original_code = f.read()
for i in range(MAX_ATTEMPTS):
try:
predict_lines, hash_to_string_dict = self.path_extractor.extract_paths(input_filename)
except ValueError as e:
print(e)
continue
results = self.model.predict(predict_lines)
prediction_results = common.parse_results(results, hash_to_string_dict, topk=SHOW_TOP_CONTEXTS)
for method_prediction in prediction_results:
print('Original name:\t' + method_prediction.original_name)
for name_prob_pair in method_prediction.predictions:
print('\t(%f) predicted: %s' % (name_prob_pair['probability'], name_prob_pair['name']))
if '|'.join(method_prediction.predictions[0]['name']) != method_prediction.original_name:
print("MATCH FOUND!", newname_of_var)
print("Tried (total:", len(closed), ") :: ", closed)
name_found = True
break
# print('Attention:')
# for attention_obj in method_prediction.attention_paths:
# print('%f\tcontext: %s,%s,%s' % (
# attention_obj['score'], attention_obj['token1'], attention_obj['path'], attention_obj['token2']))
loss, all_strings, all_grads = self.model.calc_loss_and_gradients_wrt_input(predict_lines)
indecies_of_var = np.argwhere(all_strings == newname_of_var.lower()).flatten()
grads_of_var = all_grads[indecies_of_var]
if grads_of_var.shape[0] > 0:
# print("current loss:",loss)
total_grad = np.sum(grads_of_var, axis=0)
# # words to increase loss
# top_replace_with = np.argsort(total_grad)[::-1][:5]
# result = [(i, total_grad[i], self.model.index_to_word[i]) for i in top_replace_with]
# print("words to increase loss:")
# print(result)
# words to decrease loss
# top_replace_with = np.argsort(total_grad)[:5]
similarity_to_var = self.get_similar_words(newname_of_var) #self.measureSimilarity(words_vocab,newname_of_var, "euclidean")
result = [(self.model.word_to_index[i], i, total_grad[self.model.word_to_index[i]]) for i in similarity_to_var]
result.sort(key=lambda v: (-v[2]))
print(result)
# similarity_to_var = self.measureSimilarity(words_vocab,newname_of_var, "cosine")
# resulte = [(i, self.model.index_to_word[i], similarity_to_var[i], total_grad[i]) for i in range(1, words_vocab.shape[0])]
#
# resulte.sort(key=lambda v: (v[2],-v[3]))
# select new name
for r in result:
if r[1] not in closed and r[1] != method_prediction.original_name.replace("|","")\
and r[2] > 0:
print(r)
newname_of_var = r[1]
break
else:
newname_of_var = None
if newname_of_var is None:
break
closed.append(newname_of_var)
print("rename", var_to_rename, "to", newname_of_var)
code = InteractivePredictor.rename_variable(original_code,var_to_rename,newname_of_var)
with open("input.java", "w") as f:
f.write(code)
if not name_found:
print("FAILED!")
print("Tried (total:", len(closed),") :: ", closed)
def predict(self):
input_filename = 'Input.java'
MAX_ATTEMPTS = 50
MAX_NODES_TO_OPEN = 10
print('Starting interactive prediction with mono adversarial search...')
while True:
print(
'Modify the file: "%s" and press any key when ready, or "q" / "quit" / "exit" to exit' % input_filename)
user_input = input()
if user_input.lower() in self.exit_keywords:
print('Exiting...')
return
print("select variable to rename:")
var_to_rename = newname_of_var = input()
name_found = False
opened = []
closed = []
with open(input_filename, "r") as f:
original_code = f.read()
for i in range(MAX_ATTEMPTS):
try:
predict_lines, hash_to_string_dict = self.path_extractor.extract_paths(input_filename)
except ValueError as e:
print(e)
continue
bfs = AdversarialSearcher(2, 2, self.model)
r = bfs.find_adversarial(predict_lines)
print(r)
print(timeit.timeit(lambda: bfs.find_adversarial(predict_lines), number=1000))
return
results = self.model.predict(predict_lines)
prediction_results = common.parse_results(results, hash_to_string_dict, topk=SHOW_TOP_CONTEXTS)
for method_prediction in prediction_results:
print('Original name:\t' + method_prediction.original_name)
for name_prob_pair in method_prediction.predictions:
print('\t(%f) predicted: %s' % (name_prob_pair['probability'], name_prob_pair['name']))
if '|'.join(method_prediction.predictions[0]['name']) == method_prediction.original_name:
print("MATCH FOUND!", newname_of_var)
print("Tried (total:", len(closed), ") :: ", closed)
name_found = True
break
# print('Attention:')
# for attention_obj in method_prediction.attention_paths:
# print('%f\tcontext: %s,%s,%s' % (
# attention_obj['score'], attention_obj['token1'], attention_obj['path'], attention_obj['token2']))
loss, all_strings, all_grads = self.model.calc_loss_and_gradients_wrt_input(predict_lines)
indecies_of_var = np.argwhere(all_strings == newname_of_var.lower()).flatten()
grads_of_var = all_grads[indecies_of_var]
if grads_of_var.shape[0] > 0:
# print("current loss:",loss)
total_grad = np.sum(grads_of_var, axis=0)
# # words to increase loss
# top_replace_with = np.argsort(total_grad)[::-1][:5]
# result = [(i, total_grad[i], self.model.index_to_word[i]) for i in top_replace_with]
# print("words to increase loss:")
# print(result)
# words to decrease loss
top_replace_with = np.argsort(total_grad)[:5]
result = [(i, total_grad[i], self.model.index_to_word[i]) for i in top_replace_with]
# select new name
for r in result:
if r[2] not in closed and r[2] != method_prediction.original_name.replace("|",""):
print(r)
newname_of_var = r[2]
break
else:
newname_of_var = None
if newname_of_var is None:
break
closed.append(newname_of_var)
print("rename", var_to_rename, "to", newname_of_var)
code = InteractivePredictor.rename_variable(original_code,var_to_rename,newname_of_var)
with open("input.java", "w") as f:
f.write(code)
if not name_found:
print("FAILED!")
print("Tried (total:", len(closed),") :: ", closed)
