Multi instance | BindingDB from TDC

[phalem]

Adding multi instance, I also add template file but it still didn't update here in validation. I hope also to provide text template for regression task as the one provide was for classification
Template I add :
"templates": [
"The molecule with the {SMILES__description} of {SMILES#} affect {Target_ID#} by {BindingDB_Kd#not &NULL}{BindingDB_Kd__names__adjective}.", # noqa: E501
"Based on the {SMILES__description} {SMILES#}, the molecule is {BindingDB_Kd#not &NULL}{BindingDB_Kd__names__adjective} to {Target_ID#}.", # noqa: E501
"Based on the {SMILES__description} {SMILES#}, protein sequence {Target#} the molecule is {BindingDB_Kd#not &NULL}{BindingDB_Kd__names__adjective} to {Target_ID#}.", # noqa: E501
"The {SMILES__description} {SMILES#} represents a molecule that target {Target_ID#} with {BindingDB_Kd#not &NULL} {BindingDB_Kd__names__adjective}.", # noqa: E501
# Instruction tuning text templates
]

[MicPie]

Hi @phalem !

Thank you for the PR and reaching out to me! :-)

I guess BindingDB_Kd is a float value, so you can rewrite the templates without the boolean setup to something like:

The molecule with the {SMILES__description} of {SMILES#} binds to {Target_ID#} with a {BindingDB_Kd__names__NOUN} of {BindingDB_Kd#}.

How does Target_ID look like? Is it a full molecule name or just some kind of id?

[MicPie]

Oh, and you can also branch off from the text_sampling branch in order to use the latest templating sampling setup.
In order to do so you can create a new branch based on text_sampling with git checkout -b new_branch and the use git cherry pick b8724226429e2005ee34e0e35bd9767b0eda0746 to reapply your commit to this PR branch.
Just let me know if you have time for that, or if something is unclear, etc.
I can also have a look and change the branch and the text template. :-)