Peter Hausamann

Negative RSA exponent for android_key_7_3_29 on Windows

3

For some reason, the `android_key_7_3_29` global variable has a negative exponent on my Windows machine. from gpsoauth import google, android_key_7_3_29 google.signature('[email protected]', 'asdf', android_key_7_3_29) raises: Traceback (most recent call last): File...

preprocessing.Splitter and Segmenter: allow reduce_index to be a callable.

3

It should be possible to pass a callable as `reduce_index` which takes the number of samples after transformation as an input and returns an index of this length.

[WIP] Release 0.5

1

Make sure you have: - [ ] added the necessary updates to `doc/content/whatsnew.rst` - [ ] added documentation for new features - [ ] run bumpversion and reset conda build...

The `reshapes` parameter is confusing and in reality only necessary if the number of dimensions changes. Proposal: 1. Check if number of dimensions has changed, if yes check for reshapes...

`attrs` and `name` are dropped e.g. in [`_call_fitted`](https://github.com/phausamann/sklearn-xarray/blob/0a8e61222a89e02665f444233e2bb2eb2bef7184/sklearn_xarray/common/base.py#L199).

Implement cross_val_* methods from sklearn.model_selection

2

These methods do not work for pipelines/estimators that change the number of samples, because of the `check_consistent_length` check.

A lot of the tests aren't really "unit" tests, but rather integration tests for the wrappers.

This is going to be a serious issue as of sklearn version 0.20 because `clone` will raise an exception. The only workaround is to use `wrap` with `compat=True`. This seems...

For example, the `EstimatorWrapper` calls `_transform` for each variable/DataArray whereas the transformers in the preprocessing module call `_transform` for each group and `_transform_var` for each variable/DataArray. There should be a...

Looping over the `data_vars` in `fit`/`predict`/`transform`/... should be done with `parallel`.