def run_from_config(params, *args):
if len(args) < 2:
print "No input file or test file given: %s:%s" % (args, len(args))
sys.exit(-1)
input_file = args[0]
test_file = args[1]
if len(args) > 2:
num_epochs = int(args[2])
else:
num_epochs = 1000
params["D"] = 2**params["logD"]
params["conv_layer1_nfilters"] = 2**params["log_conv_layer1_nfilters"]
params["conv_layer2_nfilters"] = 2**params["log_conv_layer2_nfilters"]
model = ConvolutionalCopyAttentionalRecurrentLearner(params)
model.train(input_file, max_epochs=num_epochs)
test_data, original_names = model.naming_data.get_data_in_recurrent_copy_convolution_format(
test_file, model.padding_size)
test_name_targets, test_code_sentences, test_code, test_target_is_unk, test_copy_vectors = test_data
eval = F1Evaluator(model)
point_suggestion_eval = eval.compute_names_f1(
test_code, original_names,
model.naming_data.all_tokens_dictionary.get_all_names())
return -point_suggestion_eval.get_f1_at_all_ranks()[1]
def run_from_config(params, *args):
if len(args) < 2:
print "No input file or test file given: %s:%s" % (args, len(args))
sys.exit(-1)
input_file = args[0]
test_file = args[1]
if len(args) > 2:
num_epochs = int(args[2])
else:
num_epochs = 1000
params["D"] = 2**params["logD"]
params["conv_layer1_nfilters"] = 2**params["log_conv_layer1_nfilters"]
params["conv_layer2_nfilters"] = 2**params["log_conv_layer2_nfilters"]
model = ConvolutionalAttentionalLearner(params)
model.train(input_file, max_epochs=num_epochs)
test_data, original_names = model.naming_data.get_data_in_convolution_format(
test_file, model.name_cx_size, model.padding_size)
test_name_targets, test_name_contexts, test_code_sentences, test_original_name_ids = test_data
ids, unique_idx = np.unique(test_original_name_ids, return_index=True)
eval = F1Evaluator(model)
point_suggestion_eval = eval.compute_names_f1(
test_code_sentences[unique_idx], original_names,
model2.naming_data.all_tokens_dictionary.get_all_names())
return -point_suggestion_eval.get_f1_at_all_ranks()[1]
def evaluate_suggestion_decisions(self, input_file):
test_data, original_names = self.naming_data.get_data_in_copy_convolution_format(input_file, self.name_cx_size, self.padding_size)
name_targets, orignal_targets, name_contexts, code_sentences, code, copy_vectors, target_is_unk, original_name_ids = test_data
code = np.array(code, dtype=np.object)
ids, unique_idx = np.unique(original_name_ids, return_index=True)
eval = F1Evaluator(self)
point_suggestion_eval = eval.compute_names_f1(code[unique_idx], original_names,
self.naming_data.all_tokens_dictionary.get_all_names())
print point_suggestion_eval
return point_suggestion_eval.get_f1_at_all_ranks()
params["train_file"] = input_file
params["test_file"] = sys.argv[4]
with ExperimentLogger("ConvolutionalCopyAttentionalRecurrentLearner",
params) as experiment_log:
model = ConvolutionalCopyAttentionalRecurrentLearner(params)
model.train(input_file, max_epochs=max_num_epochs)
model.save("copy_convolutional_att_rec_model" +
os.path.basename(params["train_file"]) + ".pkl")
model2 = ConvolutionalCopyAttentionalRecurrentLearner.load(
"copy_convolutional_att_rec_model" +
os.path.basename(params["train_file"]) + ".pkl")
test_data, original_names = model2.naming_data.get_data_in_recurrent_copy_convolution_format(
params["test_file"], model2.padding_size)
test_name_targets, test_code_sentences, test_code, test_target_is_unk, test_copy_vectors = test_data
#name_ll = model2.model.log_prob_with_targets(test_code_sentences, test_name_targets)
#print "Test name_ll=%s" % name_ll
eval = F1Evaluator(model2)
point_suggestion_eval = eval.compute_names_f1(
test_code, original_names,
model2.naming_data.all_tokens_dictionary.get_all_names())
print point_suggestion_eval
results = point_suggestion_eval.get_f1_at_all_ranks()
print results
experiment_log.record_results({
"f1_at_rank1": results[0],
"f1_at_rank5": results[1]
})