__init__.py 文件源码

def split(args):
    chunks = []

    for reads_per_bc_file, bam, gem_group in itertools.izip(args.reads_per_bc,
                                                            args.barcode_chunked_bams,
                                                            args.chunk_gem_groups):
        subsample_rate = args.subsample_rate[str(gem_group)]

        with open(reads_per_bc_file) as f:
            reads_per_bc = []
            for line in f:
                _, reads = line.strip().split()
                reads_per_bc.append(float(reads) * subsample_rate)

        max_reads = np.max(reads_per_bc + [0.0])

        # vdj_asm is hard-coded to use a maximum of 200k reads / BC.
        max_reads = min(MAX_READS_PER_BC, max_reads)

        # The assembly step takes roughly num_reads * MEM_BYTES_PER_READ bytes of memory to complete each BC.
        mem_gb = max(2.0, int(np.ceil(MEM_BYTES_PER_READ * max_reads / 1e9)))

        chunks.append({
            'chunked_bam': bam,
            'gem_group': gem_group,
            '__mem_gb': mem_gb,
        })

    # If there were no input reads, create a dummy chunk
    if not chunks:
        chunks.append({'chunked_bam': None})
    return {'chunks': chunks, 'join': {'__threads': 4}}