[MRG] Add example comparing PCR and PLS#17151
Conversation
|
@TomDLT @ogrisel @lorentzenchr @rth @cmarmo if you have time |
cmarmo
left a comment
There was a problem hiding this comment.
@NicolasHug, thanks! Learned a lot!
|
Thanks all for your reviews so far. I have updated the example according to the comments. LMK what you think |
lorentzenchr
left a comment
There was a problem hiding this comment.
LGTM (+1)
Maybe, I'm a bit pedantic about axis labeling. But it helps if you skim over the text and only look at the plots.
|
Maybe @ogrisel @glemaitre @rth can give this a quick look? It's been reviewed already! |
TomDLT
left a comment
There was a problem hiding this comment.
Great example !
I would add a line in each axes of the third plot, showing the prediction of each model. This would emphasize that this third figure is conceptually different from the second one, and that the 1D projection is actually how the model is doing the predictions.
Co-authored-by: Tom Dupré la Tour <tom.dupre-la-tour@m4x.org>
…rn into pcr_vs_pls_example
TomDLT
left a comment
There was a problem hiding this comment.
More nitpicks.
Otherwise I think we can merge !
Co-authored-by: Tom Dupré la Tour <tom.dupre-la-tour@m4x.org>
|
Nice example ! |
This is a simple example comparing Principal Component Regression and Partial Least Squares Regression. The goal is to illustrate how PLS can outperfor PCR when the target is strongly correlated with features that have a low variance (which are then dropped by PCA)
Can be merged before or after #17095 (reviews welcome there too!)