SLEAP v1.0.10a7

Pre-release of minor version update with performance tweaks and bug fixes.

Changelog

• Update to TensorFlow 2.1.2 (security patch)
• Switch to ID-based hashing for LabeledFrame. This dramatically increases the performance of frame manipulation operations.
• Several convenience methods for sleap.Labels:
  • Add describe method to Labels for easy inspection of dataset stats
  • Add has_frame method to Labels for quick checking of frame existence
  • Add remove_user_instances and remove_predictions for quick dataset cleanup
• Remove predicted instances in existing frames before merging in active learning results (fixes #413)
• Conda environment.yml clean-up: de-duplicates dependencies managed by pip
• Set h5py version requirement to 2.10.0 to prevent TensorFlow model loading issue
• Added experimental maDLC CSV labels importing support (#412)
• Keep previous zoom state when navigating across frames with fit to instances (#416)
• Add head type to the run name suffix when saving a training pipeline (#415)
• Update built-in baseline profiles
  • Remove dataset specific fields (e.g., "anchor_part")
  • Add medium/large RF variants
  • Remove unused profiles
  • Disable tensorboard logging by default
  • Standardize optimization parameters

Installing

Using Conda (Windows):
Create new environment sleap_alpha (recommended):
conda create -n sleap_alpha -c sleap/label/dev sleap=1.0.10a7
or to update inside an existing environment:
conda install -c sleap/label/dev sleap=1.0.10a7

Using PyPI (Linux/Mac):
pip install sleap==1.0.10a7