Bpnet-lite

==============

Highlights
----------

- Changed the function name from `calculate_attributions` to `attribute`
to be more in line with the `predict` and `marginalize` functions. The
functionality and usage should be the same.
- Changed the nomenclature from "interpret" to "attribute" to be more
consistent with the name of the function and what is used colloquially.

==============

Highlights
----------

- Replaced the negative sampling code with a simpler approach that only
considers bins of signal rather than operates at bp resolution. This code
is much faster and more robust but may produce slightly worse GC matches.
- The negative sampling code now allows you to pass in a bigwig so that
only regions that pass a threshold are selected.

==============

Highlights
----------

- Changed the `warning_threshold` argument to only print a warning rather
than end the process when the model exceeds it.
- Added support for plotting annotations alongside `plot_attributions`
- Fixed various minor bugs.

==============

Highlights
----------

- Changed the shape of the returned one-hot encoded sequences to match
the documentation.
- Fixed an issue with dinucleotide shuffling when not all nucleotides
are present.

==============

Highlights
----------

- Fixed an issue with ChromBPNet reading.

==============

Highlights
----------

- Added in reading of TensorFlow-formatted ChromBPNet models from the
official repo using the `from_chrombpnet` commands to the BPNet and
ChromBPNet objects.