- Support multitrack search
- Update active learning sampling strategies
- Make windows selectable
- Visualize the progress of the actively-learned classifier
- Update and expand examples to use our newly trained autoencoders
- Tons of bug fixes and performance improvements
- Represent datasets and encoders as classes rather than loose collections of
dicts - Make autoencoders optional, i.e., some tracks might be encodable (and thus searchable) but the encoding might not be decodable.
- Cache chunked, encoded, and autoencoded data as HDF5 files to avoid having to hold everything in memory.
- Replace internal scatterplot with
regl-scatterplot - Update the UI to the latest version of HiGlassApp
- Remove Bootstrap v3, which is used by HiGlass's view header, as it's not needed and imposes a security risk
- Search across a single bigWig datasets over up to a few chromosomes