def main():
data = load_data()
if ("deepsol1" in static_args.parameter_setting_id):
x_train, y_train = data['train']['src'], data['train']['tgt']
dynamic_args['num_classes'] = len(set(y_train))
x_train = np.array(utils.pad_sequecnes(x_train, static_args.maxlen))
y_train = utils.get_one_hot(y_train, dynamic_args['num_classes'])
print('Training data: ', x_train.shape)
x_test, y_test = data['test']['src'], data['test']['tgt']
x_test = np.array(utils.pad_sequecnes(x_test, static_args.maxlen))
y_test = utils.get_one_hot(y_test, dynamic_args['num_classes'])
print('Test data: ', x_test.shape)
else:
x_train, x_train_bio, y_train = data['train']['src'], data['train'][
'src_bio'], data['train']['tgt']
dynamic_args['num_classes'] = len(set(y_train))
x_train = np.array(utils.pad_sequecnes(x_train, static_args.maxlen))
x_train_bio = np.array(x_train_bio)[:, :-1]
dynamic_args['num_bio_feats'] = int(x_train_bio.shape[1])
y_train = utils.get_one_hot(y_train, dynamic_args['num_classes'])
print('Training data: ', x_train.shape)
print('Training data Bio: ', x_train_bio.shape)
print(x_train_bio[1])
x_test, x_test_bio, y_test = data['test']['src'], data['test'][
'src_bio'], data['test']['tgt']
x_test = np.array(utils.pad_sequecnes(x_test, static_args.maxlen))
x_test_bio = np.array(x_test_bio)[:, :-1]
y_test = utils.get_one_hot(y_test, dynamic_args['num_classes'])
print('Test data: ', x_test.shape)
print('Test data Bio: ', x_test_bio.shape)
best_model = utils.load_model(get_model_path())
if ("deepsol1" in static_args.parameter_setting_id):
[pred_test, pred_prob_test
] = get_classification_prediction(best_model, x_test, y_test)
else:
[pred_test, pred_prob_test
] = get_classification_prediction(best_model, [x_test, x_test_bio],
y_test)
print('Finished testing')
#save ion disk
get_classification_performance_path()
save_classification_prediction(pred_test, pred_prob_test)
def main():
data = load_data()
x_train, y_train = data['train']['src'], data['train']['tgt']
dynamic_args['num_classes'] = len(set(y_train))
x_train = np.array(utils.pad_sequecnes(x_train, static_args.maxlen))
y_train = utils.get_one_hot(y_train, dynamic_args['num_classes'])
print('Training data: ', x_train.shape)
x_val, y_val = data['valid']['src'], data['valid']['tgt']
x_val = np.array(utils.pad_sequecnes(x_val, static_args.maxlen))
y_val = utils.get_one_hot(y_val, dynamic_args['num_classes'])
print('Validation data: ', x_val.shape)
x_test, y_test = data['test']['src'], data['test']['tgt']
x_test = np.array(utils.pad_sequecnes(x_test, static_args.maxlen))
y_test = utils.get_one_hot(y_test, dynamic_args['num_classes'])
print('Test data: ', x_test.shape)
model = Models.DeepSol(static_args, dynamic_args).fetch_model_def()
model.compile(loss='binary_crossentropy',
optimizer=get_optimizer(),
metrics=['accuracy'])
print(model.summary())
# Training
model.fit(x_train,
y_train,
batch_size=dynamic_args['batch_size'],
epochs=int(static_args.epochs),
validation_data=(x_val, y_val),
callbacks=get_callbacks())
# Load the best model. Model that performs best on the validation data
best_model = utils.load_model(get_model_path())
# Calculate classification performance of the best model on the val and test
[acc_val, score_report_val, cm_val,
pred_val] = get_classification_performance(best_model, x_val, y_val)
[acc_test, score_report_test, cm_test,
pred_test] = get_classification_performance(best_model, x_test, y_test)
results_filename_with_path = get_classification_performance_path()
if os.path.exists(results_filename_with_path):
os.remove(results_filename_with_path)
# save on disk
save_classification_performance('Validation Accuracy: ', str(acc_val))
save_classification_performance('Test Accuracy: ', str(acc_test))
save_classification_performance('Score Report Test: : ',
str(score_report_test))
save_classification_performance('Confusion Matrix test: ', str(cm_test))
save_classification_prediction(pred_test)
def main():
data = load_data()
x_train, x_train_bio, y_train = data['train']['src'], data['train'][
'src_bio'], data['train']['tgt']
dynamic_args['num_classes'] = len(set(y_train))
class_weights = class_weight.compute_class_weight('balanced',
np.unique(y_train),
y_train)
x_train = np.array(utils.pad_sequecnes(x_train, static_args.maxlen))
x_train_bio = np.array(x_train_bio)[:, :-1]
dynamic_args['num_bio_feats'] = int(x_train_bio.shape[1])
y_train = utils.get_one_hot(y_train, dynamic_args['num_classes'])
print('Training data: ', x_train.shape)
print('Training data Bio: ', x_train_bio.shape)
print(x_train_bio[1])
x_val, x_val_bio, y_val = data['valid']['src'], data['valid'][
'src_bio'], data['valid']['tgt']
x_val = np.array(utils.pad_sequecnes(x_val, static_args.maxlen))
x_val_bio = np.array(x_val_bio)[:, :-1]
y_val = utils.get_one_hot(y_val, dynamic_args['num_classes'])
print('Validation data: ', x_val.shape)
print('Validation data Bio: ', x_val_bio.shape)
x_test, x_test_bio, y_test = data['test']['src'], data['test'][
'src_bio'], data['test']['tgt']
x_test = np.array(utils.pad_sequecnes(x_test, static_args.maxlen))
x_test_bio = np.array(x_test_bio)[:, :-1]
y_test = utils.get_one_hot(y_test, dynamic_args['num_classes'])
print('Test data: ', x_test.shape)
print('Test data Bio: ', x_test_bio.shape)
model = Models.DeepSol(static_args, dynamic_args).fetch_model_def()
model.compile(loss='binary_crossentropy',
optimizer=get_optimizer(),
metrics=['accuracy'])
print(model.summary())
# Training
# Either use both bio and protein feats or just one of them
acc_val, score_report_val, cm_val, pred_val = None, None, None, None
if dynamic_args['biofeats'] is not None and dynamic_args[
'protein_seq_feats'] is not None:
model.fit([x_train, x_train_bio],
y_train,
batch_size=dynamic_args['batch_size'],
class_weight="auto",
epochs=int(static_args.epochs),
validation_data=([x_val, x_val_bio], y_val),
callbacks=get_callbacks())
best_model = utils.load_model(get_model_path())
[acc_val, score_report_val, cm_val,
pred_val] = get_classification_performance(best_model,
[x_val, x_val_bio], y_val)
[acc_test, score_report_test, cm_test,
pred_test] = get_classification_performance(best_model,
[x_test, x_test_bio],
y_test)
elif dynamic_args['protein_seq_feats'] is not None:
model.fit(x_train,
y_train,
batch_size=dynamic_args['batch_size'],
class_weight="auto",
epochs=int(static_args.epochs),
validation_data=(x_val, y_val),
callbacks=get_callbacks())
best_model = utils.load_model(get_model_path())
[acc_val, score_report_val, cm_val,
pred_val] = get_classification_performance(best_model, x_val, y_val)
[acc_test, score_report_test, cm_test,
pred_test] = get_classification_performance(best_model, x_test,
y_test)
elif dynamic_args['biofeats'] is not None:
model.fit(x_train_bio,
y_train,
batch_size=dynamic_args['batch_size'],
class_weight="auto",
epochs=int(static_args.epochs),
validation_data=(x_val_bio, y_val),
callbacks=get_callbacks())
best_model = utils.load_model(get_model_path())
[acc_val, score_report_val, cm_val,
pred_val] = get_classification_performance(best_model, x_val_bio,
y_val)
[acc_test, score_report_test, cm_test,
pred_test] = get_classification_performance(best_model, x_test_bio,
y_test)
results_filename_with_path = get_classification_performance_path()
if os.path.exists(results_filename_with_path):
os.remove(results_filename_with_path)
# save on disk
save_classification_performance('Validation Accuracy: ', str(acc_val))
save_classification_performance('Test Accuracy: ', str(acc_test))
save_classification_performance('Score Report Test: : ',
str(score_report_test))
save_classification_performance('Confusion Matrix test: ', str(cm_test))
save_classification_prediction(pred_test)
def main():
data = load_data()
#To perform 10-fold cross-validation
kf = KFold(n_splits=10,shuffle=False)
if (static_args.parameter_setting_id=="deepsol1"):
x_train, y_train = data['train']['src'], data['train']['tgt']
dynamic_args['num_classes'] = len(set(y_train))
x_train = np.array(utils.pad_sequecnes(x_train, static_args.maxlen))
y_train = utils.get_one_hot(y_train, dynamic_args['num_classes'])
print('Training data: ', x_train.shape)
x_val, y_val = data['valid']['src'], data['valid']['tgt']
x_val = np.array(utils.pad_sequecnes(x_val, static_args.maxlen))
y_val = utils.get_one_hot(y_val, dynamic_args['num_classes'])
print('Valid data: ', x_val.shape)
x_full = np.concatenate((x_train,x_val),axis=0)
y_full = np.concatenate((y_train,y_val),axis=0)
print('Full Train data: ', x_full.shape)
else:
x_train, x_train_bio, y_train = data['train']['src'], data['train']['src_bio'], data['train']['tgt']
dynamic_args['num_classes'] = len(set(y_train))
x_train = np.array(utils.pad_sequecnes(x_train, static_args.maxlen))
x_train_bio = np.array(x_train_bio)[:, :-1]
dynamic_args['num_bio_feats'] = int(x_train_bio.shape[1])
y_train = utils.get_one_hot(y_train, dynamic_args['num_classes'])
print('Training data: ', x_train.shape)
print('Training data Bio: ', x_train_bio.shape)
x_val, x_val_bio, y_val = data['valid']['src'], data['valid']['src_bio'], data['valid']['tgt']
x_val = np.array(utils.pad_sequecnes(x_val, static_args.maxlen))
x_val_bio = np.array(x_val_bio)[:, :-1]
y_val = utils.get_one_hot(y_val, dynamic_args['num_classes'])
print('Valid data: ', x_val.shape)
print('Valid data Bio: ', x_val_bio.shape)
x_full = np.concatenate((x_train,x_val),axis=0)
x_full_bio = np.concatenate((x_train_bio,x_val_bio),axis=0)
y_full = np.concatenate((y_train,y_val),axis=0)
print('Full Train data: ',x_full.shape)
print('Full Train Bio: ',x_full_bio.shape)
best_model = utils.load_model(get_model_path())
print('Loaded best model for ',static_args.parameter_setting_id)
#Get file where results are to be saved
results_filename_with_path = get_classification_performance_path()
if os.path.exists(results_filename_with_path):
os.remove(results_filename_with_path)
#Keep average scores for cross_validation
acc_test_vec, mcc_test_vec = [],[]
count = 1
for train_index,test_index in kf.split(x_full):
print('Starting CV Iteration: ',str(count))
x_test = x_full[test_index]
y_test = y_full[test_index]
y_test = np.array(y_test, dtype='int32')
if (static_args.parameter_setting_id=='deepsol1'):
[acc_test, score_report_test, cm_test, mcc_test, pred_test, pred_prob_test] = get_classification_performance(best_model,x_test,y_test)
else:
x_test_bio = x_full_bio[test_index]
[acc_test, score_report_test, cm_test, mcc_test, pred_test, pred_prob_test] = get_classification_performance(best_model,[x_test,x_test_bio],y_test)
#Save output on disk
save_classification_performance('Iteration: '+str(count),'')
save_classification_performance('Test Accuracy: ',str(acc_test))
save_classification_performance('Test MCC: ',str(mcc_test))
save_classification_performance('Score Report Test: : ',str(score_report_test))
save_classification_performance('Confusion Matrix test: ',str(cm_test))
acc_test_vec.append(acc_test)
mcc_test_vec.append(mcc_test)
count=count+1
mean_acc = (1.0*sum(acc_test_vec))/len(acc_test_vec)
save_classification_performance('Mean CV accuracy: ',str(mean_acc))
mean_mcc = (1.0*sum(mcc_test_vec))/len(mcc_test_vec)
save_classification_performance('Mean CV MCC: ',str(mean_mcc))
print('Finished cross-validation')