def test():
aa_list = 'ACDEFGHIKLMNPQRSTVWY'
para_dict = {
'seq_len': 18,
'batch_size': 100,
'model_name': 'Seq_32x2_16_regres',
'epoch': 20,
'num_samples': 1000,
'learning_rate': 0.001,
'step_size': 5,
'n_filter1': 32,
'n_filter2': 64,
'filter_size1': 5,
'filter_size2': 5,
'fc_hidden_dim': 16,
'dropout_rate': 0.5,
'stride': 1
}
data, out = loader.synthetic_data(num_samples=para_dict['num_samples'],
seq_len=para_dict['seq_len'],
aa_list=aa_list,
type='regressor')
data = loader.encode_data(data)
train_loader, test_loader = loader.train_test_loader(
data, out, test_size=0.3, batch_size=para_dict['batch_size'])
model = CNNx2_regressor(para_dict)
model.fit(train_loader)
output = model.predict(test_loader)
labels = np.vstack([i for _, i in test_loader])
r2 = model.evaluate(output, labels)
def test():
aa_list = 'ACDEFGHIKLMNPQRSTVWY'
para_dict = {
'batch_size': 100,
'seq_len': 18,
'num_samples': 1000,
'model_name': 'DNN_2Layer_class',
'epoch': 20,
'learning_rate': 0.001,
'fc_hidden_dim': 32,
'dropout_rate': 0.5,
'model_type': 'classification'
}
data, out = loader.synthetic_data(num_samples=para_dict['num_samples'],
seq_len=para_dict['seq_len'],
aa_list=aa_list)
data = loader.encode_data(data)
train_loader, test_loader = loader.train_test_loader(
data, out, test_size=0.3, batch_size=para_dict['batch_size'])
model = DNN_classifier(para_dict)
model.fit(train_loader)
output = model.predict(test_loader)
labels = np.vstack([i for _, i in test_loader])
mat, acc, mcc = model.evaluate(output, labels)
def test():
aa_list = 'ACDEFGHIKLMNPQRSTVWY'
para_dict = {
'num_samples': 1000,
'batch_size': 100,
'seq_len': 18,
'model_name': 'Seq_32x1_16',
'optim_name': 'Adam',
'epoch': 20,
'learning_rate': 0.001,
'step_size': 5,
'n_filter': 32,
'filter_size': 5,
'fc_hidden_dim': 16,
'dropout_rate': 0.5
}
data, out = loader.synthetic_data(num_samples=para_dict['num_samples'],
seq_len=para_dict['seq_len'],
aa_list=aa_list)
data = loader.encode_data(data)
train_loader, test_loader = loader.train_test_loader(
data, out, test_size=0.3, batch_size=para_dict['batch_size'])
model = CNN_classifier(para_dict)
model.fit(train_loader)
output = model.predict(test_loader)
labels = np.concatenate([i for _, i in test_loader])
mat, acc, mcc = model.evaluate(output, labels)
# demo for optimization
para_json = os.path.join(model.model_path, 'train_parameters.json')
print(para_json)
para_dict = json.load(open(para_json, 'r'))
prev_model = CNN_classifier(para_dict)
prev_model.net_init()
prev_model.load_model()
seed_seqs, labels = next(iter(train_loader))
new_seqs = prev_model.optimization(seed_seqs, step_size=0.001, interval=10)
# check sequence identity
new_seqs.permute(0, 2, 1).argmax(dim=1).unsqueeze(dim=1)[0, :, :]
seed_seqs.permute(0, 2, 1).argmax(dim=1).unsqueeze(dim=1)[0, :, :]