Changelog¶
Next release¶
Added in 1.2.2¶
Install from bioconda
Support for 109 taxonomic levels in eggNOG 5 (was three before) (e.g.
deepnog infer -db eggnog5 -t 1239
for Firmicutes)Support for COG2020 (use
deepnog infer -db cog2020 -t 1
)
Fixes/changes in 1.2.2¶
Requirement PyYAML
Test class imports
Exit on requesting unavailable device (instead of raising an error)
1.2.1 - 2020-08-28¶
Added in 1.2.1¶
Training custom models: Users can now train additional models for further tax. levels of eggNOG 5 or even different orthology databases
TensorBoard status reports: Follow training/validation loss online
Support for configuration file (
deepnog_config.yml
)Model quality assessment
Changed in 1.2.1¶
The command line invocation now uses two subcommands:
deepnog train
for training new models, anddeepnog infer
for general orthologous group assignment (and model quality assessment)
Fixed in 1.2.1¶
Fixed packaging issue in 1.2.0 (which was subsequently removed altogether)
Several additional bug fixes and smaller changes
1.1.0 - 2020-02-28¶
Added¶
EggNOG5 root (tax 1) prediction
Changed¶
Package structure changed for higher modularity. This will require changes in downstream usages.
Remove network weights from the repository, because files are too large for github and/or PyPI.
deepnog
automatically downloads these from CUBE servers, and caches them locally.More robust inter-process communication in data loading
Fixes¶
Fix error on very short amino acid sequences
Fix error on unrecognized symbols in sequences (stop codons etc.)
Fix multiprocess data loading from gzipped files
Fix type mismatch in deepencoding embedding layer (Windows only)