v0.2.2

Changes

• 79811cd output SAM when provided a reference.
• 033d904 alignment summary .tsv.
• 9e35b81 grab bag of training improvements

v0.2.1

Changes

• c89cc04 summary tsv output.
• dd25dbb better calibrate qscores.
• 7a90664 simpler chunking for training data + validation split on read level.

v0.2.0

Changes

• 5a12b00 new model configuration that is both more accurate and converges faster.
• f330dba optional single read chunking and stitching for higher performance inference.
• aac9b9e initial support for faster custom CUDA runtime.
• f440ca2 initial pair decoding workflow.

v0.1.5

Changes

eab453d switch to more robust read normalising approach.

v0.1.4

Changes

bfd972d fastq output support.

Usage

$ bonito basecaller --fastq dna_r9.4.1 /data/reads > basecalls.fastq

Note: when outputting to fastq a viterbi decoder is used instead of a beam search which impacts accuracy slightly.

v0.1.2

Changes

52ccf04 hold out a set of chunks for model evaluation.
36deabc bonito download for retrieving pretrained models and training assets.
db30bdc bonito convert for chunking up hdf5 training files.
fe4dfd9 multi-gpu training support.
43327e3 resume training.
6b38f64 don't write empty fasta records.
3439bb6 store the parameters used to chunk the training data.
4567938 debian packaging.

v0.0.5

• Increased basecalling performance.