CHANGELOG

1.1.2
-----
- use a new preprocessing strategy (segmenting the genome and more effective I/O operations)
- modify data loading process
- add support for large input sequence lengths (up to 64,000 bp)
- add a new training control way (without Ray; see train_utils.py, printer_utils.py) 
- update documentation

1.1.1
-----
- fix a bug in calc_kmer_corr
- add more doc for the scaling section
- remove unused code

1.1.0
-----
- add a new network model (model 3)
- add support for AdamW optimizer
- update documentation

1.0.1
-----
- remove janggu dependency
- add rtd documentation
- delete unused code

1.0.0
-----
- update network model (add middle-scale layers)
- improve preprocessing code (numpy implementation; HDF5 related code)
- improve optimization code ('--weight_decay_auto'; '--restart_lr')
- improve documentation

0.9.1
-----
- clean unnecessary code
- update weight initialization for bias
- add more detailed documentation

0.9.0
-----
- First public version