example_training.py

# Copyright (c) Michael Mazanetz (NovaData Solutions LTD.), Silvia Amabilino (NovaData Solutions LTD.,
# University of Bristol), David Glowacki (University of Bristol). All rights reserved.
# Licensed under the GPL. See LICENSE in the project root for license information.

"""
This example shows how to train an RNN on a set of smiles and how to predict new smiles with the trained model.
This is just an example to show how to construct the model, because the example data set only contains 50 samples, which
is not enough to train the RNN.
"""

from molbot import smiles_generator, data_processing
import os
import numpy as np

# Reading the data
data_dir = os.path.join("..", "data")
data_path = os.path.join(data_dir, "example_data_2.csv")
in_d = open(data_path, 'r')

# Parsing the data
molecules = []

for line in in_d:
    line = line.rstrip()
    molecules.append(line)

print("The total number of molecules is: %i \n" % (len(molecules)))

# One-hot encode the molecules
dp = data_processing.Molecules_processing()
X = dp.onehot_encode(molecules)
# y is the same as X, but shifted by one character to the left and with the last character equal to the padding 'A' character
idx_A = dp.char_to_idx['A']
y = np.zeros(X.shape)
y[:, :-1, :] = X[:, 1:, :]
y[:, -1, idx_A] = 1

# Creating the model
estimator = smiles_generator.Smiles_generator(epochs=20, batch_size=100, tensorboard=False, hidden_neurons_1=100,
                                              hidden_neurons_2=100, dropout_1=0.3, dropout_2=0.5, learning_rate=0.001,
                                              validation=0.01)

# Training the model on the one-hot encoded molecules
estimator.fit(X, y)

# Predicting 10 new molecules from the fitted model at a temperature of 0.75
X_pred_hot = dp.get_empty(10)
pred_hot = estimator.predict(X_pred_hot, temperature=0.75)
pred = dp.onehot_decode(pred_hot)

# Print some predicted SMILES (they will be nonsense here because the model is trained on very few samples)
for smile in pred:
    print(smile)

# Saving the estimator for later re-use
estimator.save("example-model.h5")
dp.save("example-dp.pickle")