[ghc-steering-committee] #380 GHC2021: Principal Component Analysis

[Richard Eisenberg]

I think a bit of unresolved discussion here is the question of whether it is the intended fate of extensions to become on-by-default, or whether we expect some extensions to stay on as a way of forcing users to opt in. In this light, my proximity to Iavor is not that surprising, as I believe we both think that some extensions are best kept as extensions. On the other hand, I believe Arnaud (joined by Simon Marlow, if I recall) has advocated most strongly that each extension should either graduate to be on-by-default or has somehow failed to pass muster. (I won't quite say that it should be removed.) This is an interesting philosophical discussion, and I think it would be good to continue, but perhaps after this round is complete.

Thanks for the analysis!

Richard

> On Jan 10, 2021, at 5:21 PM, Joachim Breitner <mail at joachim-breitner.de> wrote:
> 
> Hi,
> 
> as a little Sunday evening statistics and visualization nerding
> relaxation, I tried to see if I can visualize where the committee
> members stand on the various kinds of extensions. It seems that
> “Principal Component Analysis” might be a good fit.
> 
> In the ~100 dimensional space where each of our ballots inhabits one
> point, it identifies the primary (and secondary) direction along which
> we are most diverse (I think). One nice thing is that this produces a
> two dimensional chart.
> 
> I was _hoping_ to maybe see “heavy type hackery” and “lots of syntactic
> sugar” emerge as the two main directions.
> 
> Well, let’s look at pca.png first…
> 
> Not quite surprising, we see Tom relatively isolated (Tom voted for 7
> extensions that nobody else voted for, and was in general very
> liberal).
> 
> Also Iavor is somewhat far off, also not surprising, given his
> reluctance to go all in on GADTs/TypeFamilies etc. Surprisingly,
> Richard is quite close to him!
> 
> Simon PJ and me are very close to each other, and somewhere in the
> middle, and the others seem to have quite an opposite stand from Iavor
> in the second principal components?
> 
> 
> Can we understand these two directions? They are essentially (I
> believe) eigenvectors in the ~100-dimensional space. I tried to
> visualize them by another scatter plot, see vectors.png. There, we can
> see that one’s vote for StarIsType contributes a lot towards the
> primary principal component, and one for Unicode a lot towards the
> secondary.
> 
> We see GADTs and TypeFamilies on the far bottom, and FunDeps not far.
> This indicates that the y-axis really might be the “heavy type level
> programming” axis (with type level programming features causing
> negative numbers).
> 
> I can’t quite make sense of the x-axis though…
> 
> 
> I stopped short of coloring the points in this graph by “extension
> cateogry”…
> 
> 
> So, no big conclusions to be drawn from this. Also, I am definitely no
> expert in the Data Sciences, so this may be misguided, or simply wrong.
> If someone wants to give it a better shot, please do! I am happy to
> provide the raw data in a simple accessible form.
> 
> 
> Cheers,
> Joachim
> 
> -- 
> Joachim Breitner
>  mail at joachim-breitner.de
>  http://www.joachim-breitner.de/
> 
> <pca.png><vectors.png>_______________________________________________
> ghc-steering-committee mailing list
> ghc-steering-committee at haskell.org
> https://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-steering-committee