Hi everybody.
I just started a PR on common testing and would really like your feedback:
https://github.com/scikit-learn/scikit-learn/pull/893
The idea is to get all estimators in sklearn and do some testing based
on the mixins.
This is still in a very early stage and there are some problems which
I'd like to discuss:
- Not all Estimator classes were default-constructable. I changed that,
for example in
the SparsePCA. I'm not sure whether defaults are sensible but it
makes testing
much easier and it might also help users (maybe?).
- Some classes are "base classes" that should not be instantiated (like
BaseLibSVM).
But these classes are not really marked as such. Some classes are
abstract base
classes, but not all of them. For example BaseLibSVM has no
abstract method.
For these, I made "__init__" an abstract method. This also prevents
users from
instantiating them by accident. Do you think this is a good idea?
- Some classes are meta estimators, like one-vs-rest, grid search,
pipeline, ensemble.
These take a base estimator class as required argument. I think
having a separate
mixin for these meta estimators would be a good idea.
As I said, any feedback would be very welcome!
Cheers,
Andy
------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and
threat landscape has changed and how IT managers can respond. Discussions
will include endpoint security, mobile security and the latest in malware
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
