Open huyajing0 opened 2 months ago
The reconstruction process is separated as a single process by default. So it is correct to first generate t-SMILES sequences using GPT/LSTM or other generation models. And then perform reconstruction to convert t-SMILES sequences into SMILES. This way, it's easy to use the shared generation model. On the other hand, the decode_single() method could be called directly to easily integrate the reconstruction process with generation models.
Hi, regarding the evaluation process mentioned in the article, I have some doubts. Is it first to use the GPT generation model to generate t-SMILES sequences, and then to reconstruct molecules based on these sequences? Is that correct?Looking forward to your reply.