def getChunkLength(args, chromSize):
"""
There's no point in parsing the GTF time over and over again needlessly.
Emprically, it seems that adding ~4x the number of workers is ideal, since
coverage is non-uniform. This is a heuristic way of approximating that.
Note that if there are MANY small contigs and a few large ones (e.g., the
max and median lengths are >10x different, then it's best to take a
different tack.
"""
if args.region:
chromSize, region_start, region_end, genomeChunkLength = getUserRegion(chromSize, args.region)
rv = np.ceil((region_start - region_end) / float(4 * args.numberOfProcessors)).astype(int)
return max(1, rv)
bl = None
if args.blackListFileName:
bl = GTF(args.blackListFileName)
lengths = []
for k, v in chromSize:
regs = blSubtract(bl, k, [0, v])
for reg in regs:
lengths.append(reg[1] - reg[0])
if len(lengths) >= 4 * args.numberOfProcessors:
rv = np.median(lengths).astype(int)
# In cases like dm6 or GRCh38, there are a LOT of really small contigs, which will cause the median to be small and performance to tank
if np.max(lengths) >= 10 * rv:
rv = np.ceil(np.sum(lengths) / (4.0 * args.numberOfProcessors)).astype(int)
else:
rv = np.ceil(np.sum(lengths) / (4.0 * args.numberOfProcessors)).astype(int)
return max(1, rv)
def getChunkLength(args, chromSize):
"""
There's no point in parsing the GTF time over and over again needlessly.
Emprically, it seems that adding ~4x the number of workers is ideal, since
coverage is non-uniform. This is a heuristic way of approximating that.
Note that if there are MANY small contigs and a few large ones (e.g., the
max and median lengths are >10x different, then it's best to take a
different tack.
"""
if args.region:
chromSize, region_start, region_end, genomeChunkLength = getUserRegion(chromSize, args.region)
rv = np.ceil((region_start - region_end) / float(4 * args.numberOfProcessors)).astype(int)
return max(1, rv)
bl = None
if args.blackListFileName:
bl = GTF(args.blackListFileName)
lengths = []
for k, v in chromSize:
regs = blSubtract(bl, k, [0, v])
for reg in regs:
lengths.append(reg[1] - reg[0])
if len(lengths) >= 4 * args.numberOfProcessors:
rv = np.median(lengths).astype(int)
# In cases like dm6 or GRCh38, there are a LOT of really small contigs, which will cause the median to be small and performance to tank
if np.max(lengths) >= 10 * rv:
rv = np.ceil(np.sum(lengths) / (4.0 * args.numberOfProcessors)).astype(int)
else:
rv = np.ceil(np.sum(lengths) / (4.0 * args.numberOfProcessors)).astype(int)
return max(1, rv)