Jacob Schreiber

We should probably aim to set up a GitHub action in the style of this one which does linting and runs unit tests on a variety of Python versions and...

From those errors, I'm not sure what's happening. I'd need to have access to the underlying data. Are your attributions the full hypothetical contributions?

Can you provide the command you're using to run it?

Thanks. I'll take a look soon. How much faster is it? Can you add in the documentation a reference to the code in single_cell_toolkit?

Thanks for looking into it. Any of the fragments files under `scATAC_clusters.zip` should be good for testing: https://zenodo.org/records/8313962

Where is this issue arising?

Fixed

The BPNet model itself should work on any length. Pay attention to the `trimming` parameter, which specifies the difference between the input length and the output length (or, specifically, half...

If you set trimming to 0 you should be able to use the same input and output length. However, you'll likely get worse predictions on the flanks because they can't...

Usually, the bias model is given far less context so that it does not inadvertently learn complex rules. We usually use a model with four layers, so the residual layers...