kwgoodman

@adament In order to test the idea of using float64 intermediate values, I need a unit test that demonstrates the problem. Does anyone have one? It may take me a...

If it works for neither 32 nor 64 bit then I don't think I can use it for a unit test (?)

Numpy and pandas use the name data type to mean float64, int32, etc. So I picked the name region to represent train, validation, etc. Plus it is easier to type....

The benefits of this proposal are especially good once #20 is done.

When we take up this proposal we may as well take the zscore of the predictions (self.y_id_sorted) to save about 3 seconds per submission when there are 1000 active submissions:...

If you store larrys in hdf5 then the label order is preserved across save and loads. Can you use hdf5 in your use case?

Sounds like a reasonable feature to add. So we'd add a sort_label (True by default) to fromlist and fromtuples. Plus update docstrings, manual, and add unit tests. Are you interested...

Take your time, I won't be able to look at it for the next couple of weeks.

That sounds good.

There is currently a swapaxes. But a rollaxis sounds like a good addition. la includes the numpy license so we can copy and paste np.rollaxis and then add label support...