def _split_data(x_data, y_data, split=0.9):
'''Split data.'''
x_data_rand, y_data_rand = theanets_utils.randomise_order(x_data, y_data)
# Split data into training and classifying:
ind = int(split * len(x_data_rand))
x_train = x_data_rand[:ind]
y_train = [[y] for y in y_data_rand[:ind]]
x_val = x_data_rand[ind:]
y_val = y_data_rand[ind:]
return x_train, y_train, x_val, y_val
def _learn(sequences, activities):
'''Attempt to learn sequence / activity relationship.'''
# Convert sequences to inputs, based on amino acid properties:
x_data = sequence_utils.get_aa_props(sequences)
x_data, y_data = theanets_utils.randomise_order(x_data, activities)
# Split data into training and classifying:
ind = int(0.8 * len(x_data))
y_train = [[y] for y in y_data[:ind]]
regressor = theanets_utils.Regressor(x_data[:ind], y_train)
regressor.train(hidden_layers=[1024])
y_pred = regressor.predict(x_data[ind:])
return regressor, y_data[ind:], y_pred
