0.2.5

• Fix a dependency conflict between keras and h5py.

0.2.4

• Sequences shorter than 4 nucleotides would break the program in the k-mer frequency step (as RNAsamba uses 2, 3 and 4-mers). Now, those sequences are ignored.

0.2.3

• The input FASTA file is now read only once.
• Several unnecessary loops were removed.

0.2.2

• The longest_orf function is now written in Rust and is ~50 times faster.
• The ORF extraction step is now parallelized.

0.2.1

• The k-mer frequency function is now written in Rust and is parallelized.

0.2.0

• The count_kmers function is now written in Rust and is approximately 5.5× faster than the old native Python implementation.

0.1.6

• Class and function names in the miniigloo module were improved.
• Updated Keras version requirements.

0.1.5

• CLI now uses subparsers for the classification and training modules.
• Changed version requirements for numpy (<1.17) to avoid unnecessary warnings.
• Disabled Tensorflow's excessive logging and warnings.
• Step number is now shown.

0.1.4

• Replace U with T in nucleotide sequences.

0.1.2

• Set the minimal Python version to 3.6
• Add the partial-length weights file
• Remove tf.logging.set_verbosity(tf.logging.ERROR), which was causing warnings in TensorFlow 1.14