def reproduce(seed,
samples_path=None,
metrics_path=None,
n_jobs=1,
device='cpu',
verbose=False,
samples=30000):
train = moses.get_dataset('train')
model = CombinatorialGenerator(n_jobs=n_jobs)
if verbose:
print("Training...")
model.fit(train)
if verbose:
print(f"Sampling for seed {seed}")
seeds = list(range((seed - 1) * samples, seed * samples))
samples = mapper(n_jobs)(model.generate_one, seeds)
if samples_path is not None:
with open(samples_path, 'w') as f:
f.write('SMILES\n')
for sample in samples:
f.write(sample + '\n')
if verbose:
print(f"Computing metrics for seed {seed}")
metrics = moses.get_all_metrics(samples, n_jobs=n_jobs, device=device)
if metrics_path is not None:
with open(metrics_path, 'w') as f:
for key, value in metrics.items():
f.write("%s,%f\n" % (key, value))
return samples, metrics
def reproduce(seed, samples_path=None, metrics_path=None,
n_jobs=1, device='cpu', verbose=False,
samples=30000):
data = moses.get_dataset('train')[:100000]
if verbose:
print("Training...")
model = HMM(n_jobs=n_jobs, seed=seed, verbose=verbose)
model.fit(data)
np.random.seed(seed)
if verbose:
print(f"Sampling for seed {seed}")
np.random.seed(seed)
samples = [model.generate_one()
for _ in range(samples)]
if samples_path is not None:
with open(samples_path, 'w') as f:
f.write('SMILES\n')
for sample in samples:
f.write(sample+'\n')
if verbose:
print(f"Computing metrics for seed {seed}")
metrics = moses.get_all_metrics(
samples, n_jobs=n_jobs, device=device)
if metrics_path is not None:
with open(samples_path, 'w') as f:
for key, value in metrics.items():
f.write("%s,%f\n" % (key, value))
return samples, metrics
def reproduce(seed, samples_path=None, metrics_path=None,
n_jobs=1, device='cpu', verbose=False,
samples=30000):
data = moses.get_dataset('train')
model = NGram(10, verbose=verbose)
model.fit(data)
np.random.seed(seed)
smiles = model.generate(samples, l_smooth=0.01)
metrics = moses.get_all_metrics(smiles, n_jobs=n_jobs, device=device)
if samples_path is not None:
with open(samples_path, 'w') as out:
out.write('SMILES\n')
for s in smiles:
out.write(s+'\n')
if metrics_path is not None:
with open(metrics_path, 'w') as out:
for key, value in metrics.items():
out.write("%s,%f\n" % (key, value))
return smiles, metrics
torch.cuda.set_device(0)
moses_qed_props_model_path = "../models/moses/"
model.load(moses_qed_props_model_path)
model.cuda()
import random
from rdkit import RDLogger
RDLogger.DisableLog('rdApp.*')
generated = []
verbose_lim = 10000
print("Sampling smiles", flush=True)
while len(generated) < 30000:
sampled = model.sample(1000)
sampled_valid = [s for s in sampled if get_mol(s)]
generated += sampled_valid
n_generated = len(generated)
if n_generated >= verbose_lim:
print("Generated %d of %d SMILES" % (len(generated), 30000), flush=True)
verbose_lim += 10000
with open("../moses_sampling/sampled_smiles.csv", "w") as f:
f.writelines("%s\n" % sm for sm in generated)
print("Calculating Metrics", flush=True)
metrics = moses.get_all_metrics(generated)
pickle.dump( metrics, open( "metrics.pkl", "wb" ) )
import os
import sys
import argparse
if __name__=='__main__':
parser = argparse.ArgumentParser()
parser.add_argument('-i', "--generated_samples", help="Nbr to generate", type=str, default='data/gen.txt')
args, _ = parser.parse_known_args()
# =======================================
script_dir = os.path.dirname(os.path.realpath(__file__))
sys.path.append(os.path.join(script_dir, 'dataloaders'))
sys.path.append(os.path.join(script_dir, 'data_processing'))
with open(os.path.join(script_dir,'..',args.generated_samples), 'r') as f :
smiles_list = [line.rstrip() for line in f]
print(f'> Read {len(smiles_list)} smiles in data/gen.txt. Computing metrics...')
metrics = moses.get_all_metrics(smiles_list)
print('MOSES benchmark metrics :')
for k,v in metrics.items():
print(k,':', f'{v:.4f}')
# to copy values to excel sheet with benchmarks
for k,v in metrics.items():
print( f'{v:.4f}')