Changelog

v0.5.0

bug fixes

change feature QUAL to SNratio, didn’t work though?

hid ‘–n_vocab’ and ‘–n_embed’ in model params

add attbigru2s2/attbilstm2s2 models

update new version of pretrained model params

v0.4.1

handle ValueError when fetching bam items in call_freqb module

change pytorch max version to 1.12.1

using logging instead of print/sys.stderr.write

v0.4.0

optimieze call_mods module, faster generating of modbam file

v0.3.4

replace ‘numpy.float’ by ‘float” in extract_features and call_mods modules

v0.3.3

more robust operation for MM/ML tags

update the call_mods model (v1 to v2), use shared params for both rnn and attention

update aggregate model (v2 to v2p)

v0.3.2

fix bug (0-pos CG in reverse strand) of call_freqb module

add output check at the start of call_mods/call_freqb

update aggregate model

v0.3.1

optimize call_mods module

add aggregate mode in call_freqb module

release a stable call_mods model and a stable call_freqb aggregate model

v0.3.0

add multi-threads support for reading .bam by pysam

force mod prob discrete integer in [0, 255], which means mod prob in [0, 1), in ML tag

fix align_hifi module, enable minimap2 and bwa

update requirements

v0.2.3

more options in train module

multi-gpu support in call_mods and trainm modules

add denoise module

Note: skip 0.2.2, cause 0.2.2 has been deleted in pypi (https://pypi.org/manage/project/ccsmeth/history/)

v0.2.1

minor fixes

change default mode in extract module

improvements in call_freqb module

v0.2.0

use hifi reads instead of subreads as input

modbam support

use (chrom, pos, strand) as key instead of (chrom, pos) to handle CG mismatch in CCS read when call_freq

v0.1.2

add pbmm2 as default aligner

enable gzip output

v0.1.1

bug fixes

v0.1.0

initialize