morgancthomas/smiles-rnn

Repository for SMILES-based RNNs for reinforcement learning-based de novo molecule generation

smiles-rnn's Issues

Patch to require i/o/s arguments

In the course of other work, I see that I made the following changes locally in train_prior.py. Would this be a good change to include?

-    required.add_argument('-i', '--train_smiles', type=str, help='Path to smiles file')
-    required.add_argument('-o', '--output_directory', type=str, help='Output directory to save model')
-    required.add_argument('-s', '--suffix', type=str, help='Suffix to name files')
+    required.add_argument('-i', '--train_smiles', type=str, help='Path to smiles file', required=True)
+    required.add_argument('-o', '--output_directory', type=str, help='Output directory to save model', required=True)
+    required.add_argument('-s', '--suffix', type=str, help='Suffix to name files', required=True)

UnboundLocalError: local variable 'input_vector' referenced before assignment

In the model files (rnn.py, transformer.py and gated_transformer.py), there is a mistake on a line causing an error:
UnboundLocalError: local variable 'input_vector' referenced before assignment

I think
input_vector = input_vector.to(self.device)
should be
input_vectors = input_vectors.to(self.device)

Error while running reinforcement.py script

Hi,

Running the reinforcement_learning.py code, the following error occurred.

(molscore) medicina@medicina:SMILES-RNN$ python reinforcement_learning.py -p Prior_GuacaMol_reprod_lstm_Epoch-10.ckpt -m ret_sim_config.json --model RNN RV2
Found existing directory, appending current time to distinguish
Device set to cuda
0%| | 0/250 [00:00<?, ?it/s]
Traceback (most recent call last):
File "/media/medicina/SMILES-RNN/model/RL.py", line 84, in _score
scores = self.scoring_function(smiles)
File "/media/medicina/SMILES-RNN/molscore/manager.py", line 586, in call
self.run_scoring_functions(smiles=smiles_to_process, file_names=file_names)
File "/media/medicina/SMILES-RNN/molscore/manager.py", line 248, in run_scoring_functions
results = function(smiles=smiles, directory=self.save_dir, file_names=file_names)
File "/media/medicina/SMILES-RNN/molscore/scoring_functions/similarity.py", line 102, in call
with Pool(self.n_jobs) as pool:
File "/home/medicina/anaconda3/envs/molscore/lib/python3.6/multiprocessing/context.py", line 119, in Pool
context=self.get_context())
File "/home/medicina/anaconda3/envs/molscore/lib/python3.6/multiprocessing/pool.py", line 174, in init
self._repopulate_pool()
File "/home/medicina/anaconda3/envs/molscore/lib/python3.6/multiprocessing/pool.py", line 229, in _repopulate_pool
for i in range(self._processes - len(self._pool)):
TypeError: 'float' object cannot be interpreted as an integer

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "reinforcement_learning.py", line 157, in
main(args)
File "reinforcement_learning.py", line 49, in main
record = RL.train(n_steps=args.n_steps, save_freq=args.save_freq)
File "/media/medicina/SMILES-RNN/model/RL.py", line 50, in train
self._train_step(step=step)
File "/media/medicina/SMILES-RNN/model/RL.py", line 187, in _train_step
scores = self._score(smiles, step)
File "/media/medicina/SMILES-RNN/model/RL.py", line 93, in _score
self.scoring_function.write_scores()
File "/media/medicina/SMILES-RNN/molscore/manager.py", line 415, in write_scores
temp = self.main_df.copy()
AttributeError: 'NoneType' object has no attribute 'copy'

What kind of error is this? Please help.

Thanks

The script `train_prior.py` does not write the training times

In the script script/train_prior.py, it crashes when writing the training times to the json file

Incompatible with Ampere Cards

Thanks for the code!

We would like to run this on an A100, however, on the Ampere cards, it requires CUDA SDK or toolkit to be v11.0<. However, this is pinned in environment.yml with cudatoolkit=10.2.89.

Is there a chance this could be updated for the latest cards?