Alexander Clausen

It fails on my system with `cupy.cuda.cufft.CuFFTError: CUFFT_EXEC_FAILED`: ``` src/libertem/api.py:707: in run_udf results = UDFRunner([udf]).run_for_dataset( src/libertem/udf/base.py:1113: in run_for_dataset for part_results, task in executor.run_tasks(tasks, cancel_id): src/libertem/executor/inline.py:23: in run_tasks result = task()...

Also note comment from @fwin22: >Some minor suggestions: It could be helpful if `exclude_pixels` in `correct` supports both a list of defect pixels, i.e., coordinates, as well as a boolean...

Also may still need some test stability fixes: ``` ___________________________________________________________________________________ test_mask_patch ___________________________________________________________________________________ @pytest.mark.with_numba def test_mask_patch(): for i in range(10): print(f"Loop number {i}") num_nav_dims = np.random.choice([1, 2, 3]) num_sig_dims = np.random.choice([2,...

Some more notes: * `Context.run_udf` should take a `Corrections` object, which can contain different parts (dark, gain, hot/cold pixels; later affine transformation matrix or similar?) * `DataSet` should have a...

Discussion: some parts (dictating slices for partitions from the outside) would be very useful for #1264, so I'll extract these into a separate PR

As an update, the GSOC project did happen: https://blogs.python-gsoc.org/en/peterbell10s-blog/ That work should be released as part of scipy 1.4 AFAIK; when that happens we can fix this issue.

We should also think about providing a "registry"-like interface, where people can register any third-party package as fft backend - that way people can also use mkl fft, which has...

Nowadays, we also have CUDA/cupy support, which needs to be taken into account when selecting the right FFT functions - for UDFs, we could offer a similar API as `UDF.xp`...

Idea: style the apply button differently depending on whether the parameters are "dirty" or not

> The notebook contains the parameters from the last run FYI with #1172 having landed, this situation changed, at least for the CoM analysis: parameters are updated as they are...