def getMates(seq, seqs, th, fdis, matches, lim = -1, failsTH = 30) :
cans = [0]*len(seqs)
cclust.counts(seq, matches, cans)
lseq = len(seq)
scans = [(-x/min(lseq,len(seqs[k])), k) for k,x in enumerate(cans) if x > 0]
heapq.heapify(scans)
fails = 0
matched = []
distances = []
for cn,k in hiter(scans) :
if lim > 0 and len(matched) >= lim:
break
djk = fdis(seq, k)
if djk <= th :
matched.append(k)
distances.append(djk)
fails = 0
else :
fails += 1
if fails > failsTH:
break
return matched,distances
def getMates(seq, seqs, lseqs, th, fdis, matches, lim = -1, failsTH = 30) :
fails = 0
matched = []
distances = []
q = cclust.counts(seq, matches, len(seqs), lseqs)
for cn,k in hiternew(q) :
if lim > 0 and len(matched) >= lim:
break
djk = fdis(seq, k)
if djk <= th :
matched.append(k)
distances.append(djk)
fails = 0
else :
fails += 1
if fails > failsTH:
break
return matched,distances
def getPotentials(iSeq, seqs, lseqs, th, fdis, matches, elem2grps, rs = None) :
seq = seqs[iSeq]
stats = True
## cclust.counts does this counting faster
##
## cans = [0]*len(seqs)
## for i in range(len(seq)-11) :
## p = seq[i:i+11]
## ms = matches.get(p)
## if ms :
## for m in ms:
## cans[m] += 1
## cans[iSeq] = 0
cans = [0]*len(seqs)
cclust.counts(seq, matches, cans)
cans[iSeq] = 0
lseq = len(seq)
# min is expensive (seriously! the volume of calls is staggering)
# len too, that is why we pre-compute them.
scans = [(-x/(lseq if lseq <= lseqs[k] else lseqs[k]), k)
for k,x in enumerate(cans) if x > 0]
# heap instead of full sorting, we usually need only a very small fraction of
# the elements. Should move to faster C code. heapq is pure python
heapq.heapify(scans)
if rs:
pfails,pok = 0,0
tries = 0
# Sequences we do not need to consider
alreadyMatched = set()
fails = 0
matched = []
distances = []
# should we make the q sub-order on length (same nmatches, bring shorter one first)
# Or another way to handle multiples from longer sequences?
for cn,k in hiter(scans):
if k in alreadyMatched :
continue
tries += 1
djk = fdis(iSeq,k)
if djk <= th :
matched.append(k)
distances.append(djk)
fails = 0
g = elem2grps.get(k)
if g :
for x in g :
alreadyMatched.add(x)
if rs:
pok = cans[k]/min(lseqs[k],lseq)
if pok < rs[0] :
rs[0] = pok
rs[1] += 1
else :
if rs and fails == 0 :
pfails = cans[k]/min(lseqs[k],lseq)
# rs[0] minimal accepted rs[1] n-accepted rs[2] total fails, rs[3] n-fails
fails += 1
# One mismatch can cause such a drop. dont give up before that.
# heigher scores are possible due to multiple matchings.
if cans[k] <= lseq-21 :
if fails > _failsHardTH :
break
if fails > 1 and rs and rs[1] > 500 and rs[3] > 500 :
p = cans[k]/min(lseqs[k],lseq)
if p < .99*rs[0] and p < (rs[2]/rs[3] + rs[0])/2 :
break
if rs :
rs[2] += pfails
rs[3] += 1
return matched,distances, (fails,tries,len(scans))