Re: Does Weka do assumption check?

> On Aug 26, 2023, at 12:02 PM, fmohr <[email protected]> wrote:
> 
> Hi all,
> 
> interesting remarks. I think that assumption checking is particularly reasonable if you, for whatever reason, expel dct that you have to make predictions *outside* the range of training data (sort of extrapolation) or if you have relatively little data. I think that there are quite some scenarios, like these, where you can obtain somehow reasonable results on training (and maybe also validation) data, but things could badly wrong on new data. This can justify the assumption checks (one could also do it post-hoc, just in case the AutoML tool tells you that LR was the best algorithm, just to be sure).
> 
> But, and in agreement with Eibe, I personally think that a reasonable EDA should be part of the analysis workflow anyway and *always* (and this I say as an AutoML guy). And if you do an EDA, you will already have covered the assumptions (depending of course on how exhaustive you went in your EDA).
> 
> By 5 cents,
> Felix
> 

Looking at the link closer I think 3 of the assumptions were based on residuals which would make them post-hoc?

As I remember there are occasions when regression can get seriously thrown off but I think they are rare. 

I think mostly you would find the first assumption off where there really isn’t a linear relationship and you could fairly easily find some other classifier that would out-perform. Which AutoML would probably find?



_______________________________________________
Wekalist mailing list -- [email protected]
Send posts to [email protected]
To unsubscribe send an email to [email protected]
To subscribe, unsubscribe, etc., visit https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html