def reward_aromatic_rings(smiles):
'''
A simple reward to encourage larger molecule length
:param smiles: list of strings
:return: reward, list of float
'''
return np.array([-1 if num is None else num+0.5 for num in num_aromatic_rings(smiles)])
def reward_aromatic_rings(smiles):
'''
A simple reward to encourage larger molecule length
:param smiles: list of strings
:return: reward, list of float
'''
if not len(smiles):
return -1 # an empty string is invalid for our purposes
atoms = num_aromatic_rings(smiles)
return [-1 if num is None else num + 0.5 for num in atoms]
sa_thresh=0,
normalize_scores=True)
max_steps = 277 #settings['max_seq_length']
def second_score(smiles):
pre_scores = 3 * 2.5 + scorer.get_scores(smiles)[0]
score = np.power(pre_scores.prod(1), 0.333)
for i in range(len(score)):
if np.isnan(score[i]):
score[i] = -1
return score
reward_fun = lambda x: scorer(x) + np.array(
[-5 * max(x - 5, 0) for x in num_aromatic_rings(x)]
) # - 0.2*np.array([0 if num is None else num for num in num_aromatic_rings(x)])# + reward_aliphatic_rings(x)# + 0.05*reward_aromatic_rings(x)#lambda x: reward_aromatic_rings(x)# #lambda x: reward_aromatic_rings(x)#
model, fitter1, fitter2 = train_policy_gradient(
molecules,
grammar,
EPOCHS=100,
BATCH_SIZE=batch_size,
reward_fun_on=reward_fun,
max_steps=max_steps,
lr_off=0.0,
lr_on=1e-4,
drop_rate=drop_rate,
decoder_type='attention', #'random',#
plot_prefix='anchor cycle',
dashboard='anchor wweak sa', #None,#
