dotnet / machinelearning

ML.NET is an open source and cross-platform machine learning framework for .NET.
https://dot.net/ml
MIT License
9.01k stars 1.88k forks source link

Scrubbing `Projection` transformers #2831

Closed Ivanidzo4ka closed 5 years ago

Ivanidzo4ka commented 5 years ago

Sub task of #2827 RFF, LpNorm, GcNorm, PCA, Whiten, Normalize. Projection catalog is bad name, and we probably should toss around this transforms across other catalogs.

wschin commented 5 years ago

They are in projection because they all map feature vectors to a low-rank space. Why do you think this name is bad?

Ivanidzo4ka commented 5 years ago

Well, for example LpNorm and GcNorm doesn't lower rank at all AFAIK. And if I correctly remember @TomFinley comment: "What is transformer if not projection of data?" Maybe DimensionalityReduction is a better catalog name than Projection

wschin commented 5 years ago

Just updated/created some catalogs. We now create Normalization for all normalizer, DimensionReduction for PCA and VectorWhiten, Kernel for RFF kernel approximation.