hyper_parameters = None
for key in ['learning_rate', 'beta', 'gamma', 'dense_sizes']:
if args.__dict__[key] == None:
if hyper_parameters == None:
tuning_file = os.path.join(ROOT_DIR, 'experiments', 'tuning', 'results', 'smad', args.antipattern, args.test_system + '.csv')
hyper_parameters = detection_utils.get_optimal_hyperparameters(tuning_file)
args.__dict__[key] = hyper_parameters[' '.join(key.split('_')).capitalize()]
# Remove the test system from the training set
systems = data_utils.getSystems()
systems.remove(args.test_system)
systems = list(systems)
# Get training data
x_train = reduce(lambda x1, x2: np.concatenate((x1, x2), axis=0), [detection_utils.getInstances(args.antipattern, s) for s in systems])
y_train = reduce(lambda x1, x2: np.concatenate((x1, x2), axis=0), [detection_utils.getLabels(args.antipattern, s) for s in systems])
# Get test data
x_test = detection_utils.getInstances(args.antipattern, args.test_system)
y_test = detection_utils.getLabels(args.antipattern, args.test_system)
# Create model
model = md.SMAD(
shape=args.dense_sizes,
input_size=x_train.shape[-1])
# To save and restore a trained model
saver = tf.train.Saver(max_to_keep=args.n_net)
# Train several neural networks
all_losses_train = []
else:
core_metrics[system] = hist_cm.getFECoreMetrics(system)
params = np.arange(0.0, 20.0, 0.5)/100. if args.antipattern == 'god_class' else np.arange(100., 300., 5)/100.
# Initialize progressbar
bar = progressbar.ProgressBar(maxval=len(params), \
widgets=['Tuning HIST parameters for ' + args.test_system + ': ' ,progressbar.Percentage()])
bar.start()
perfs = []
for i, alpha in enumerate(params):
bar.update(i)
overall_prediction = np.empty(shape=[0, 1])
overall_labels = np.empty(shape=[0, 1])
for system in systems:
smells = [entityName for entityName, ratio in core_metrics[system].items() if ratio[0]>alpha]
prediction = detection_utils.predictFromDetect(args.antipattern, system, smells)
overall_prediction = np.concatenate((overall_prediction, prediction), axis=0)
overall_labels = np.concatenate((overall_labels, detection_utils.getLabels(args.antipattern, system)), axis=0)
perfs.append(detection_utils.mcc(overall_prediction, overall_labels))
bar.finish()
output_file_path = os.path.join(ROOT_DIR, 'experiments', 'tuning', 'results', 'hist', args.antipattern, args.test_system + '.csv')
indexes = np.argsort(np.array(perfs))
with open(output_file_path, 'w') as file:
file.write("Alpha;MCC\n")
for i in reversed(indexes):
file.write("{0:.3f};{1}\n".format(params[i], perfs[i]))
session.run(model.learning_step, feed_dict=feed_dict_train)
if __name__ == "__main__":
args = parse_args()
# Remove the test system from the set of systems
systems = data_utils.getSystems()
systems.remove(args.test_system)
# Store instances and labels for each system
instances = {}
labels = {}
for system in systems:
instances[system] = detection_utils.getInstances(args.antipattern, system, True)
labels[system] = detection_utils.getLabels(args.antipattern, system)
# Initialize progress bar
bar = progressbar.ProgressBar(maxval=args.n_test, \
widgets=['Performing cross validation for ' + args.test_system + ': ' ,progressbar.Percentage()])
bar.start()
output_file_path = os.path.join(ROOT_DIR, 'experiments', 'tuning', 'results', 'smad', args.antipattern, args.test_system + '.csv')
params = []
perfs = []
for i in range(args.n_test):
learning_rate, beta, gamma, dense_sizes = generateRandomHyperParameters()
params.append([learning_rate, beta, gamma, dense_sizes])
pred_overall = np.empty(shape=[0, 1])
import numpy as np
systems = data_utils.getSystems()
overall_prediction_incode = np.empty(shape=[0, 1])
overall_prediction_hist = np.empty(shape=[0, 1])
overall_prediction_jd = np.empty(shape=[0, 1])
overall_prediction_vote = np.empty(shape=[0, 1])
overall_prediction_asci = np.empty(shape=[0, 1])
overall_prediction_smad = np.empty(shape=[0, 1])
overall_labels = np.empty(shape=[0, 1])
for system in systems:
# Get labels
labels = detection_utils.getLabels('feature_envy', system)
overall_labels = np.concatenate((overall_labels, labels), axis=0)
# Compute performances for InCode
prediction_incode = incode.predict(system)
overall_prediction_incode = np.concatenate(
(overall_prediction_incode, prediction_incode), axis=0)
# Compute performances for HIST
prediction_hist = hist_fe.predict(system)
overall_prediction_hist = np.concatenate(
(overall_prediction_hist, prediction_hist), axis=0)
# Compute performances for JDeodorant
prediction_jd = jdeodorant_fe.predict(system)
overall_prediction_jd = np.concatenate(
import numpy as np
systems = data_utils.getSystems()
overall_prediction_decor = np.empty(shape=[0, 1])
overall_prediction_hist = np.empty(shape=[0, 1])
overall_prediction_jd = np.empty(shape=[0, 1])
overall_prediction_vote = np.empty(shape=[0, 1])
overall_prediction_asci = np.empty(shape=[0, 1])
overall_prediction_smad = np.empty(shape=[0, 1])
overall_labels = np.empty(shape=[0, 1])
for system in systems:
# Get labels
labels = detection_utils.getLabels('god_class', system)
overall_labels = np.concatenate((overall_labels, labels), axis=0)
# Compute performances for DECOR
prediction_decor = decor.predict(system)
overall_prediction_decor = np.concatenate(
(overall_prediction_decor, prediction_decor), axis=0)
# Compute performances for HIST
prediction_hist = hist_gc.predict(system)
overall_prediction_hist = np.concatenate(
(overall_prediction_hist, prediction_hist), axis=0)
# Compute performances for JDeodorant
prediction_jd = jdeodorant_gc.predict(system)
overall_prediction_jd = np.concatenate(
return parser.parse_args()
if __name__ == '__main__':
args = parse_args()
# Remove the test system from the training set and build dataset
systems = data_utils.getSystems()
systems.remove(args.test_system)
systems = list(systems)
# Get overall labels
overall_labels = np.empty(shape=[0, 1])
for system in systems:
overall_labels = np.concatenate(
(overall_labels, detection_utils.getLabels('feature_envy',
system)),
axis=0)
params = [(atfd, laa, fdp) for atfd in range(1, 6) for laa in range(1, 6)
for fdp in range(1, 6)]
# Initialize progressbar
bar = progressbar.ProgressBar(maxval=len(params), \
widgets=['Tuning InCode parameters for ' + args.test_system + ': ' ,progressbar.Percentage()])
bar.start()
perfs = []
count = 0
for atfd, laa, fdp in params:
count += 1
bar.update(count)
# Remove the test system from the training set
systems = data_utils.getSystems()
systems.remove(args.test_system)
systems = list(systems)
# Get training instances
x_train = reduce(
lambda x1, x2: np.concatenate((x1, x2), axis=0),
[detection_utils.getInstances(args.antipattern, s) for s in systems])
# Get ASCI training labels
tools_predictions_list = [
asci_utils.get_tools_predictions(args.antipattern, s) for s in systems
]
labels_list = [
detection_utils.getLabels(args.antipattern, s) for s in systems
]
y_train_asci = asci_utils.get_asci_labels(tools_predictions_list,
labels_list)
# Get test data, note that here y_test contains the real labels while y_train_asci contains ASCI labels, i.e., tools' indexes
x_test = detection_utils.getInstances(args.antipattern, args.test_system)
y_test = detection_utils.getLabels(args.antipattern, args.test_system)
tools_predictions_test = asci_utils.get_tools_predictions(
args.antipattern, args.test_system)
# Train and compute ensemble prediction on test set
predictions = np.zeros((args.n_tree, x_test.shape[0], 1))
for i in range(args.n_tree):
clf = tree.DecisionTreeClassifier(
min_samples_split=args.min_samples_split,