Hacker News new | ask | show | jobs
Information Theoretic Metrics for Multi-Class Predictor Evaluation (tech.magnetic.com)
13 points by dcrosta 4017 days ago
1 comments

I'm not entirely clear on why this was needed - The Matthews Correlation Coefficient can easily be used in multi-class multi-label classification and regression scenarios.
I'm guessing here -- because there are highly dependent uncorrelated datasets?

It's easy to give examples:

https://en.wikipedia.org/wiki/Normally_distributed_and_uncor...

https://en.wikipedia.org/wiki/Matthews_correlation_coefficie... is defined for binary (two-class) classifications only.
MCC is trivially extended to support regression and multi-label. See the implementation in http://grey.colorado.edu/emergent