if __name__ == "__main__":
args = parser.parse_args()
configs = generate_all_model_configs(
activations=args.activation,
init_methods=args.init,
max_ic50_values=args.max_ic50,
dropout_values=args.dropout,
minibatch_sizes=args.minibatch_size,
embedding_sizes=args.embedding_size,
n_pretrain_epochs_values=args.pretrain_epochs,
n_training_epochs_values=args.training_epochs,
hidden_layer_sizes=args.hidden_layer_size,
learning_rates=args.learning_rate,
optimizers=args.optimizer)
print("Total # configurations = %d" % len(configs))
training_datasets, _ = load_data(
args.binding_data_csv_path,
max_ic50=args.max_ic50,
peptide_length=9,
binary_encoding=False)
combined_df = evaluate_model_configs(
configs=configs,
results_filename=args.output,
train_fn=lambda config: evaluate_model_config_by_cross_validation(
config,
training_datasets,
min_samples_per_allele=args.min_samples_per_allele,
cv_folds=args.cv_folds))
hyperparameter_performance(combined_df)
init=args.init,
n_pretrain_epochs=args.pretrain_epochs,
n_epochs=args.training_epochs,
dropout_probability=args.dropout,
max_ic50=args.max_ic50,
minibatch_size=args.minibatch_size,
learning_rate=args.learning_rate,
optimizer=args.optimizer)
models = [make_model(config) for _ in range(args.ensemble_size)]
binary_encoding = (args.embedding_size == 0)
training_datasets, _ = load_data(
filename=args.training_csv,
peptide_length=9,
max_ic50=args.max_ic50,
binary_encoding=binary_encoding)
X_all = np.vstack([dataset.X for dataset in training_datasets.values()])
Y_all = np.concatenate([
dataset.Y
for dataset in training_datasets.values()
])
for model in models:
model.fit(
X_all,
Y_all,
nb_epoch=args.pretrain_epochs,
batch_size=args.minibatch_size,
