rxn.onmt_models.translation.rxn_translation

rxn.onmt_models.translation.rxn_translation(src_file, tgt_file, pred_file, model, n_best, beam_size, batch_size, gpu, max_length=300, as_external_command=False)[source]

Do a forward or retro translation.

This function takes care of tokenizing/detokenizing the input. In principle, by adapting the “invalid” placeholder, this could also work when input/output are full reactions.

Note: no check is made that the source is canonical.

Parameters

  • src_file (Union[str, PathLike]) – source file (tokenized or detokenized).

  • tgt_file (Union[str, PathLike, None]) – ground truth file (tokenized or detokenized), not mandatory.

  • pred_file (Union[str, PathLike]) – file where to save the predictions.

  • model (Union[str, PathLike]) – model to do the translation

  • n_best (int) – number of predictions to make for each input.

  • beam_size (int) – beam size.

  • batch_size (int) – batch size.

  • gpu (bool) – whether to use the GPU.

  • max_length (int, default: 300) – maximum sequence length.

  • as_external_command (bool, default: False) – runs the onmt command instead of Python code.

Return type

None