def sample(self, n):
"""Sample a batch size n of experience"""
if len(self.memory) < n:
raise IndexError('Size of memory ({}) is less than requested sample ({})'.format(len(self), n))
else:
scores = [x[1] for x in self.memory]
sample = np.random.choice(len(self), size=n, replace=False, p=scores / np.sum(scores))
sample = [self.memory[i] for i in sample]
smiles = [x[0] for x in sample]
scores = [x[1] for x in sample]
prior_likelihood = [x[2] for x in sample]
tokenized = [self.voc.tokenize(smile) for smile in smiles]
encoded = [self.voc.encode(tokenized_i) for tokenized_i in tokenized]
encoded = Dataset.collate_fn(encoded)
return encoded, np.array(scores), np.array(prior_likelihood)
def initiate_from_file(self, fname, scoring_function, Prior):
"""Adds experience from a file with SMILES
Needs a scoring function and an RNN to score the sequences"""
with open(fname, 'r') as f:
smiles = []
for line in f:
smile = line.split()[0]
if Chem.MolFromSmiles(smile):
smile = Chem.MolToSmiles(Chem.MolFromSmiles(smile), isomericSmiles=False)
smiles.append(smile)
scores = scoring_function(smiles)
tokenized = [self.voc.tokenize(smile) for smile in smiles]
encoded = [self.voc.encode(tokenized_i) for tokenized_i in tokenized]
encoded = Dataset.collate_fn(encoded)
prior_likelihood, _ = Prior.likelihood(encoded.long())
prior_likelihood = prior_likelihood.data.cpu().numpy()
new_experience = zip(smiles, scores, prior_likelihood)
self.add_experience(new_experience)
