crf-beam

Beam search for Bonito CRF models.

Usage

>>> import torch
>>> from crf_beam import beam_search
>>>
>>> T, C, S = 1000, 320, 64
>>> scores = torch.rand(T, C)
>>> back_guides = torch.rand(T + 1, S)
>>> post = torch.rand(T, S)
>>>
>>> sequence, qstring, moves = beam_search(scores, back_guides, post)
>>> sequence
'ACACTCGCAGCGCGATACGACTGATCGAGATATACTCAGTGTACACAGT...'
>>> qstring
'(&''(****95444"@732"55",--/5555"5"@@E"E4""","-#**...'

Developer Quickstart

Python

$ python3 -m venv venv3
$ source venv3/bin/activate
(venv3) $ pip install -r requirements.txt
(venv3) $ make build
(venv3) $ make test

WASM

$ make wasm
$ node tests/test.js

import { crf_beam_decoder, initModule } from '@nanopore/crf-beam'

let Module = await initModule();

const decodedData = await crf_beam_decoder(Module, {
    scores: new Float32Array(arrayBuffer),
    numberSamplesInRead,
    modelStride,
    features,
});