Matthew Bryan

👋 Chiming in from the LiberTEM side, both dates are good for me, with a slight preference for the Thursday afternoon.

Another note, I pushed a prototype branch with the capacity to allocated chunked, cupy-backed Dask arrays as buffers: https://github.com/matbryan52/LiberTEM/compare/83179d9..delayed_cupy One thing I hadn't quite appreciated is that even when a...

/azp run libertem.libertem-data

> Sorry for the delay - I'm currently reviewing pending PRs that should maybe land before #1207 and I think this is one of them. > > > This touches...

This should possibly just be added to #1207 , I didn't see how wide in scope that was!

:+1: I think it would fit in with #1270 if the int `cudas` will assign workers round-robin to the available GPUs (so `cudas=4` on a two-GPU machine would create 2...

> If the partitioning as done currently makes things hard, we can think about changing that completely (for example by removing some degrees of freedom in the `DataSet`/`Partition` interface). For...

The PR has been updated with a prototype `DataSet` implementation and an accompanying notebook. https://github.com/LiberTEM/LiberTEM/blob/29130f66702f2ffdee256172245136e62393fc67/prototypes/timepix/TimePix3_DataSet.ipynb I'm copying the last cell here to talk about its limitations: - The `hits =>...

This is not a LiberTEM-specific issue but I put it here for visibility: there is also a bug that can occur in some setups (Windows + Numpy/Intel MKL particularly) that...

I don't know where the master version of this checklist is stored, but we should add the necessary conda-related steps to it once that's in place.