def test_natural_cutoff(self):
lo, hi = 0.2, 0.5
self.assertAlmostEqual(
sequt.natural_cutoff([0.28, 0.40, 0.50, 0.65], lo, hi), 0.28)
self.assertAlmostEqual(
sequt.natural_cutoff([0.1, 0.17, 0.21, 0.30, 0.51], lo, hi), 0.21)
self.assertAlmostEqual(
sequt.natural_cutoff([0.1, 0.17, 0.21, 0.28, 0.37, 0.45, 0.51], lo,
hi), 0.2)
def test_natural_cutoff (self):
lo, hi = 0.2, 0.5
self.assertAlmostEqual(
sequt.natural_cutoff(
[0.28, 0.40, 0.50, 0.65],
lo, hi
),
0.28
)
self.assertAlmostEqual(
sequt.natural_cutoff(
[0.1, 0.17, 0.21, 0.30, 0.51],
lo, hi
),
0.21
)
self.assertAlmostEqual(
sequt.natural_cutoff(
[0.1, 0.17, 0.21, 0.28, 0.37, 0.45, 0.51],
lo, hi
),
0.2
)
def cluster_by_jaccard (self, minlen=5): #, threshold=0.7):
"""
"""
lass = self.lass
dim = len(lass)
jaccdist = self.jaccarrays['distance']
for i, istr in enumerate(lass):
if len(istr) < minlen:
continue
threshold = sequt.natural_cutoff(jaccdist[i], lo=0.2, hi=0.3)
for j in xrange(dim):
jstr = lass[j]
if (i == j) or len(jstr) < minlen:
continue
if jaccdist[i,j] <= threshold:
self.clusters.merge(istr, jstr)
def link_across_length_levels (self): #, threshold=0.75):
""" Link across all length levels to create larger clusters.
"""
#assert 0 < threshold <= 1
#threshold = 0.3
lass = self.lass
dim = len(lass)
for i, istr in enumerate(lass):
vec_i = self.editdist[i] # array1d
dummy = np.ones(dim) * 999
for j in xrange(dim):
if i == j: continue
dummy[j] = 1.0 * vec_i[j] / min(len(istr), len(lass[j]))
jmin = np.argmin(dummy)
threshold = sequt.natural_cutoff(dummy, lo=0.2, hi=0.4)
if dummy[jmin] <= threshold:
self.clusters.merge(istr, lass[jmin])
def merge_within_length_level (self):
""" Find out small clusters within each length level.
"""
#assert 0 < threshold <= 1
lass = self.lass
for length, lo_hi in lass.lenbounds.iteritems():
lo, hi = lo_hi
if hi - lo < 2:
continue
#for i in xrange(lo, hi-1):
# for j in xrange(i+1, hi):
# if self.editdist[i, j] <= length * 0.5:
# self.clusters.merge(lass[i], lass[j])
for i in xrange(lo, hi):
# Find the best cut-off in [0.2, 0.5]
threshold = sequt.natural_cutoff(
self.editdist[i, lo:hi] /float(length), lo=0.2, hi=0.4)
for j in xrange(lo, hi):
if i != j and self.editdist[i, j] <= threshold * length:
self.clusters.merge(lass[i], lass[j])