Update to 0.3.0 breaks example

[jhnnsrs]

Create utility functions that I rely on a daily basis!

The 0.3.0 update breaks the listed example though:
Seems like the api for the reduction function has changed.

from xarray_multiscale import multiscale
import numpy as np

data = np.arange(4)
x = multiscale(data, np.mean, (2,))
print(x)

Traceback (most recent call last):
  File "tt.py", line 5, in <module>
    x = multiscale(data, np.mean, (2,))
  File "/usr/local/lib/python3.8/site-packages/xarray_multiscale/multiscale.py", line 131, in multiscale
    downscaled = downscale(source, reduction, scale, pad_mode=pad_mode)
  File "/usr/local/lib/python3.8/site-packages/xarray_multiscale/multiscale.py", line 294, in downscale
    downscaled_data = downscale_dask(
  File "/usr/local/lib/python3.8/site-packages/xarray_multiscale/multiscale.py", line 278, in downscale_dask
    meta = reduction(
  File "<__array_function__ internals>", line 5, in mean
  File "/usr/local/lib/python3.8/site-packages/numpy/core/fromnumeric.py", line 3440, in mean
    return _methods._mean(a, axis=axis, dtype=dtype,
  File "/usr/local/lib/python3.8/site-packages/numpy/core/_methods.py", line 167, in _mean
    rcount = _count_reduce_items(arr, axis, keepdims=keepdims, where=where)
  File "/usr/local/lib/python3.8/site-packages/numpy/core/_methods.py", line 76, in _count_reduce_items
    items *= arr.shape[mu.normalize_axis_index(ax, arr.ndim)]
numpy.AxisError: axis 2 is out of bounds for array of dimension 1

pip dump:
dask==2021.11.2 numcodecs==0.9.1 numpy==1.21.4 s3fs==2021.11.1 xarray==0.17.0 xarray-multiscale==0.3.0 zarr==2.8.3

Example still working in 0.2.0 :)
Best

[d-v-b]

Yes, I'm sorry about that. (I thought I was the only user.... )

For reductions I was using the dask corasen method directly. This method takes a function like np.mean and wraps it in functionality that accomplishes windowed averaging. However, working with functions that don't have the same API as np.mean is tough with this approach. So now, multiscale takes reduction functions that directly do windowed downscaling, where the first argument is an N-dimensional array, and the second argument is a tuple of scale factors.

I included two reduction functions that work in the new style here: https://github.com/JaneliaSciComp/xarray-multiscale/blob/main/src/xarray_multiscale/reducers.py#L6

I can update the documentation to reflect the new way of doing things. And thanks for opening the issue!

[jhnnsrs]

A typed well documented scientific library? You are certainly not the only user ;)

I was able to find the reduction method and it works like a charm once you modify

def downscale(
    array: DataArray,
    reduction: Callable[[NDArray[Any], Tuple[int, ...]], NDArray[Any]],
    scale_factors: Sequence[int],
    pad_mode: str,
    **kwargs: Any,
) -> Any:

    to_downscale = normalize_array(array, scale_factors, pad_mode=pad_mode)
    downscaled_data = downscale_dask(
        to_downscale.data, reduction, scale_factors, **kwargs
    )
    downscaled_coords = downscale_coords(to_downscale, scale_factors)
    return DataArray(
        downscaled_data, downscaled_coords, attrs=array.attrs, **dims=array.dims**
    )

The original code just infered the dims from the coords, which according to a beautiful error was deprecated
and is no longer supported for this xarray (0.17.0) version. Do you want me to right a pull request?

Thanks for maintainig :)

[d-v-b]

yes, a PR would be greatly appreciated!

[jhnnsrs]

#11 here it is ;)

[d-v-b]

fixed in #11