Likelihood of sampling external SMILES

[Horizon371]

Hello,

I have a list of SMILES that were not generated by REINVENT. I want to check the likelihood of sampling them from a trained agent. Do you know how I could do that?

Thank you!

Best regards,
Cristian

[halx]

Hi,

you would have to write a small script to do that yourself.

Cheers,
Hannes.

[Horizon371]

Would this be correct?

import torch
from REINVENT4.reinvent.models.reinvent.models.model import Model
import numpy as np

def compute_likelihood(model_path, smiles:list):
    model_dict = torch.load(model_path)
    model = Model.create_from_dict(model_dict, "inference", torch.device("cpu"))
    likelihoods = model.likelihood_smiles(smiles)
    print(np.mean(likelihoods.data.numpy()))

[halx]

For Reinvent that looks fine but for the other generators you would have to go through ModelAdapter().