Clarify confusion regarding feature filtering #140
Added ability to set random seed for CLI and sets fixed random seeds for qiime2 #101
Correcting matching between metadata and biom table and clarifying the min-feature-count parameter #99
Added Tensorboard's HParams functionality to standalone #95
Changed the summary-interval parameter to accept floating-point values #108
Filtered out FutureWarnings caused by TensorFlow/Tensorbard, filtered out all RuntimeWarnings, and added a silent option that suppresses progress bar information and all TensorFlow warnings #106
Various README updates about comparing your model to null/baseline models, specifying formulas, Q2 scores, etc. #103, #109
Pinned TensorFlow's version to be at least 1.15 and less than 2. This is due in part to recent security issues that have been identified in TensorFlow (1, 2) -- the latter of these does not seem to have been addressed yet on conda, but neither of these issues should have an impact on normal usage of Songbird.
- Pinning the TensorFlow version in this way may be a temporary solution, depending on how we move forward from here.
- #109
Enable duplicate metadata ids #89
Added Q2 statistic explanation and documentation #87
Added compatibility with the new qiime2 differential type #60
Minor release bump. Differential types are now inherited from q2 types.
Minor release bump. qiime2 snares are resolved.
Initial alpha release. Multinomial regression API, standalone command line interface and qiime2 interface should be stable.