Running chebifier locally

[arnaucoma24]

Dear Chebifier team,

First of all, thank you very much for this amazing and useful resource!

I have a quick question regarding the implementation of Chebifier on my local machine. How should we prepare the config file to obtain the same results as on the web server? We are currently using the file attached below and just wanted to confirm whether it is fine not to specify the model weights.

Thanks in advance!

electra:
    load_model: electra_chebi50_v241
resgated:
    load_model: resgated_chebi50_v241
chemlog_peptides:
    type: chemlog_peptides
chemlog_element:
    type: chemlog_element
chemlog_organox:
    type: chemlog_organox
c3p:
    load_model: c3p_with_weights
chebi_lookup:
    type: chebi_lookup

[arnaucoma24]

Following on my previous comment, which ensemble_type, use_confidence and resolve_inconsistencies flags should we use through the CLI?

Thanks a lot!

[sfluegel05]

Hi,

first of all, thanks for your interest and sorry for the late reply.

The configuration above should give you the same models that are currently used on the web server. The only thing missing are the weights for chemlog and the chebi_lookup. chemlog gets a model_weight: 100 and chebi_lookup gets a model_weight: 10. See also this config file for reference (the only thing missing here is the ChEBI lookup): https://github.com/ChEB-AI/python-chebifier/blob/dev/chebifier/ensemble.yml

For the hyperparameters, I would recommend sticking to the default values (ensemble_type="wmv-f1", use_confidence=True, resolve_inconsistencies=True`).

For reference, here is the config file that is currently used by Chebifier (the referenced files can be found on huggingface, exact file names might differ slightly):
Chebifier_config.json

I hope this helps.