- Added a filter for genes that are partially or fully predicted in the padding region.
- Introduced an option to specify the minimum length of input sequences used for predictions. This significantly reduces runtime for input files containing many short sequences that do not require gene structure prediction.
Bugfixes:
- Hmm parallel_factor can be at most seq_len/2.
- Fixed write_tf_records for very short sequences.
- Fixed automatic loading of training weights when running Tiberius without softmasking.