On 10/17/2017 11:18 PM, Ismael Lemhadri wrote:
How about editing the various chunks of code concerned to add the
option to scale the parameters, and set it by default to NOT scale?
This would make what happens clear without the redundancy Andreas
mentioned, and would add more convenience to the user shall they want
to scale their data.
I don't feel that would add a lot, and it would still requires the users
to read the docs.
There are many ways to scale, and applying any of them is very easy with
scikit-learn.
The main source of confusion seems to be that you expected the PCA to
scale, and it doesn't.
It doesn't say anywhere that it scales, and scaling is not part of the
definition of PCA
(in contrast to subtracting the mean).
I guess part of the confusion came from the somewhat cryptic docstring
about SVD, but you fixed that.
_______________________________________________
scikit-learn mailing list
scikit-learn@python.org
https://mail.python.org/mailman/listinfo/scikit-learn
