Implementation of one-vs-one approach

edwinb-ai / LeastSquaresSVM

A Least Squares Support Vector Machine implementation in pure Julia

https://edwinb-ai.github.io/LeastSquaresSVM/dev/

MIT License

1 stars 1 forks source link

Implementation of one-vs-one approach #26

Closed edwinb-ai closed 3 years ago

edwinb-ai commented 3 years ago

The 1-vs-1 approach creates several binary classification SVMs and utilizes a voting scheme. This is the one that Sci-kit Learn uses.

By using scitypes of the kind Multiclass, a one-vs-one approach should be easy to implement.

edwinb-ai commented 3 years ago

The use of scitypes is not really the issue here. This just means that one must type-coerce the correct target column. The voting scheme is the difficult part. A tie-break scheme is still needed.