neulab / ExplainaBoard

Interpretable Evaluation for AI Systems
MIT License
361 stars 36 forks source link

Custom Analysis Todo List #500

Open qjiang002 opened 2 years ago

qjiang002 commented 2 years ago

Custom features

Now users can specify custom features in either the output json file or a global config file. Potential improvements are:

Custom functions

[unsupported] we also aim to allow users to specify feature function so that the analysis could be performed more flexibly.

odashi commented 2 years ago

@qjiang002 Good point. Moreover, the FileLoaderMetadata mechanism reads the entire output file to load the metadata, which is obviously redundant. Decoupling the metadata file from the output may also improve this efficiency. https://github.com/neulab/ExplainaBoard/blob/37145588a75b34cc9b9f5feb82a76f23bf692c95/explainaboard/loaders/file_loader.py#L177