def run_addcarbon(smi_train, results_base, seed=None):
if seed is not None:
np.random.seed(seed)
# setup results directory
dir_results = Path(f'{results_base}/{timestamp()}')
os.makedirs(dir_results)
np.random.seed(seed)
add_carbon = AddCarbonSampler(smi_train)
sampled_smiles = add_carbon.generate(10000)
fn_sampled = str(dir_results / 'addcarbon_smiles.txt')
with open(fn_sampled, 'w') as f:
f.write('\n'.join(sampled_smiles))
fn_guacamol_results = str(dir_results / 'guacamol_results.json')
assess_distribution_learning(add_carbon,
smi_train,
json_output_file=fn_guacamol_results)
with open(fn_guacamol_results) as f:
results = json.load(f)
for b in results['results']:
print(f"{b['benchmark_name']}: {b['score']}")
print(f'Saved results in: {dir_results}')
def main(params):
# setup random seeds
set_seed(params.seed)
params.ar = True
if not os.path.isdir(params.output_dir): os.makedirs(params.output_dir)
print("Loading the model from {0}".format(params.model_path))
# load everything from checkpoint
model_params, dico, model = reload_ar_checkpoint(params.model_path)
if params.local_cpu is False:
model = model.cuda()
# evaluate distributional results
generator = SmilesTransformerGenerator(params, dico, model,
params.sample_temperature)
json_file_path = os.path.join(params.output_dir,
'distribution_learning_results.json')
smiles_output_path = os.path.join(params.output_dir,
'generated_smiles.txt')
print("Starting distributional evaluation")
t1 = time.time()
if params.evaluate is True:
assess_distribution_learning(generator,
chembl_training_file=params.dist_file,
json_output_file=json_file_path,
benchmark_version=params.suite)
else:
smiles_list = generator.generate(params.num_samples)
with open(smiles_output_path, 'w') as f:
for smiles in smiles_list:
f.write(smiles + '\n')
t2 = time.time()
print("Total time taken {}".format(t2 - t1))
def main():
parser = argparse.ArgumentParser()
parser.add_argument(
'--smiles_file',
help=
'Location of the ChEMBL dataset to use for the distribution benchmarks.',
default='data/guacamol_v1_all.smiles')
parser.add_argument(
'--pickle_directory',
help=
'Directory containing pickle files with the distribution statistics',
default=None)
parser.add_argument('--seed', type=int, default=0)
parser.add_argument('--n_jobs', type=int, default=-1)
parser.add_argument('--generations', type=int, default=1000)
parser.add_argument('--population_size', type=int, default=100)
parser.add_argument('--num_sims', type=int, default=40)
parser.add_argument('--max_children', type=int, default=25)
parser.add_argument('--max_atoms', type=int, default=60)
parser.add_argument('--init_smiles', type=str, default='CC')
parser.add_argument('--random_start', action='store_true')
parser.add_argument('--output_dir', type=str, default=None)
parser.add_argument('--suite', default='v2')
args = parser.parse_args()
if args.output_dir is None:
args.output_dir = os.path.dirname(os.path.realpath(__file__))
if args.pickle_directory is None:
args.pickle_directory = os.path.dirname(os.path.realpath(__file__))
np.random.seed(args.seed)
setup_default_logger()
# save command line args
with open(
os.path.join(args.output_dir, 'distribution_learning_params.json'),
'w') as jf:
json.dump(vars(args), jf, sort_keys=True, indent=4)
sampler = GB_MCTS_Sampler(pickle_directory=args.pickle_directory,
n_jobs=args.n_jobs,
random_start=args.random_start,
num_sims=args.num_sims,
max_children=args.max_children,
init_smiles=args.init_smiles,
max_atoms=args.max_atoms,
generations=args.generations,
population_size=args.population_size)
json_file_path = os.path.join(args.output_dir,
'distribution_learning_results.json')
assess_distribution_learning(sampler,
json_output_file=json_file_path,
chembl_training_file=args.smiles_file,
benchmark_version=args.suite)
def main(config):
setup_default_logger()
set_seed(config.seed)
generator = OrganGenerator(config)
json_file_path = os.path.join(config.output_dir,
'distribution_learning_results.json')
assess_distribution_learning(generator,
chembl_training_file=config.dist_file,
json_output_file=json_file_path,
benchmark_version=config.suite)
def main(config):
setup_default_logger()
set_seed(config.seed)
if config.output_dir is None:
config.output_dir = os.path.dirname(os.path.realpath(__file__))
generator = VaeGenerator(config)
json_file_path = os.path.join(config.output_dir, 'distribution_learning_results.json')
assess_distribution_learning(generator,
chembl_training_file=config.dist_file,
json_output_file=json_file_path,
benchmark_version=config.suite)
def main(params):
# setup random seeds
set_seed(params.seed)
params.ar = True
print("Loading the model from {0}".format(params.model_path))
# load everything from checkpoint
model_params, dico, model = reload_ar_checkpoint(params.model_path)
if params.local_cpu is False:
model = model.cuda()
# evaluate distributional results
generator = SmilesTransformerGenerator(params, dico, model)
json_file_path = os.path.join('/'.join(params.model_path.split('/')[:-1]),
'distribution_learning_results.json')
print("Starting distributional evaluation")
t1 = time.time()
assess_distribution_learning(generator,
chembl_training_file=params.dist_file,
json_output_file=json_file_path,
benchmark_version=params.suite)
t2 = time.time()
print("Total time taken {}".format(t2 - t1))
parser = argparse.ArgumentParser(description='Distribution learning benchmark for SMILES RNN',
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument('--seed', default=42, type=int, help='Random seed')
parser.add_argument('--model_path', default=None, help='Full path to SMILES RNN model')
parser.add_argument('--output_dir', default=None, help='Output directory')
parser.add_argument('--dist_file', default='data/guacamol_v1_all.smiles', help='Distribution file')
args = parser.parse_args()
device = 'cuda' if torch.cuda.is_available() else 'cpu'
logger.info(f'device:\t{device}')
set_random_seed(args.seed, device)
if args.output_dir is None:
args.output_dir = os.path.dirname(os.path.realpath(__file__))
if args.model_path is None:
dir_path = os.path.dirname(os.path.realpath(__file__))
args.model_path = os.path.join(dir_path, 'pretrained_model', 'model_final_0.473.pt')
model_def = Path(args.model_path).with_suffix('.json')
model = load_rnn_model(model_def, args.model_path, device, copy_to_cpu=True)
generator = SmilesRnnGenerator(model=model, device=device)
json_file_path = os.path.join(args.output_dir, 'distribution_learning_results.json')
assess_distribution_learning(generator,
chembl_training_file=args.dist_file,
json_output_file=json_file_path)
parser.add_argument('--suite', default='v2')
args = parser.parse_args()
device = 'cuda' if torch.cuda.is_available() else 'cpu'
logger.info(f'device:\t{device}')
set_random_seed(args.seed, device)
if args.output_dir is None:
args.output_dir = os.path.dirname(os.path.realpath(__file__))
if args.model_path is None:
dir_path = os.path.dirname(os.path.realpath(__file__))
args.model_path = os.path.join(dir_path, 'pretrained_model',
'model_final_0.473.pt')
model_def = Path(args.model_path).with_suffix('.json')
model = load_rnn_model(model_def,
args.model_path,
device,
copy_to_cpu=True)
generator = SmilesRnnGenerator(model=model, device=device)
json_file_path = os.path.join(args.output_dir,
'distribution_learning_results.json')
assess_distribution_learning(generator,
chembl_training_file=args.dist_file,
json_output_file=json_file_path,
benchmark_version=args.suite)
from guacamol.assess_distribution_learning import assess_distribution_learning
from generative_playground.molecules.guacamol_utils import DummyMoleculeGenerator
from generative_playground.molecules.lean_settings import get_data_location
# # naive sampling
# train_mols = get_data_location(source="ChEMBL:train")['source_data']
# my_gen = DummyMoleculeGenerator('distribution_naive_smiles.zip', maximize_reward=False)
# assess_distribution_learning(my_gen, train_mols)
# stable-ish run with discriminator sampling
#
files = [
['distribution_naive_no_priors_smiles.zip', 'distribution_naive_no_priors_smiles_2.zip'],
'distribution_naive_uncond_priors_smiles.zip', # is the one with unconditional priors
'distribution_naive_smiles.zip', # is the one with conditional priors
'distribution_discr_eps0.2_smiles.zip']
train_mols = get_data_location(source="ChEMBL:train")['source_data']
my_gen = DummyMoleculeGenerator(files[0], # eps0.2_smiles.zip',
maximize_reward=False,
keep_last=1e4)
assess_distribution_learning(my_gen, train_mols)