alankrantas/sefr_multiclass_classifier

Implementations of multiclass version of SEFR linear-time fast classifier (TinyML)

The chances of finding out what’s really going on in the universe are so remote, the only thing to do is hang the sense of it and keep yourself occupied.

--- The Hitchhiker's Guide to the Galaxy

Name's Alan Wang👋

A rare hybrid of a software engineer slash technical/marketing writer with translator background and have variuos knowledge in frontend/backend languages, embedded programming/Maker projects and machine learning. Currently based in Tapipei, Taiwan.

• Visit my homepage or my résumé for more personal information.
• Currently the writer, developer and maintainer of LOC Documentation, the official user guide for FST Network's serverless data product platform, Logic Operating Centre.

sefr_multiclass_classifier's Issues

Small mistake in bias calculation

Hey, in the Python code (sefr.py) the bias calculation does the following:

bias = -(neg_labels.size * pos_score_avg + # calculate weighted average of bias
         pos_indices.size * neg_score_avg) / (neg_labels.size + pos_indices.size)

However, I think you meant to do:

bias = -(neg_indices.size * pos_score_avg + # calculate weighted average of bias
         pos_indices.size * neg_score_avg) / (neg_indices.size + pos_indices.size)

Instead of neg_labels, use neg_indices. The reason is that neg_labels.size is the length of the entire dataset but neg_indices.size is the number of negative examples.