Investigate Whether `label_number` is Used or Obsolete in ChEBI Dataset Class

[aditya0by0]

Description

We need to understand whether the label_number property in the ChEBIOverX class and its derivatives (ChEBIOver100, ChEBIOver50) is actually used anywhere in the dataset creation process or pipeline.

1. Is label_number referenced or used in any part of the codebase, including dataset creation, processing, or any downstream tasks?

   • If used, what purpose does it serve?

2. If label_number is not used, can it be considered obsolete, and should it be removed to clean up the codebase?

3. Assess whether retaining label_number serves any potential future purpose, or if it should be refactored.

Relevant Code

class ChEBIOverX(_ChEBIDataExtractor):
    LABEL_INDEX: int = 3
    SMILES_INDEX: int = 2
    READER: dr.ChemDataReader = dr.ChemDataReader
    THRESHOLD: int = None

    @property
    def label_number(self) -> int:
        return 854

    def select_classes(self, g: nx.Graph, split_name: str, *args, **kwargs) -> List:
        # Implementation...
        return nodes


class ChEBIOver100(ChEBIOverX):
    THRESHOLD: int = 100

    def label_number(self) -> int:
        return 854


class ChEBIOver50(ChEBIOverX):
    THRESHOLD: int = 50

    def label_number(self) -> int:
        return 1332

[sfluegel05]

As far as I am aware, this property is obsolete. @MGlauer You probably added it - do you know what the purpose was?

[aditya0by0]

@sfluegel05
I believe that the label_number property was originally introduced to hardcode the number of labels for each specific dataset class, allowing it to be passed to the model as the out_dim parameter via link_arguments in the Lightning CLI. This was intended to eliminate the need for users to manually specify model.out_dim in the CLI when running chebai.

However, this property was never actually linked to out_dim or utilized (evident through code), and users have been manually providing model.out_dim through the CLI for each run.

In PR #74 for Issue #73, changes have been made to dynamically determine num_of_labels from the DataModule and pass it to the model. Since label_number was never used and is now redundant, I think we can remove it as part of the same PR.

[sfluegel05]

That makes sense. Please add that to the PR.