def test_class2tokens(self):
classes = 'T-VKTV-R'
tokens = 'tʰ ɔ x t ə r'.split(' ')
out = class2tokens(classes, tokens)
_ = class2tokens([['T'], ['-VKTV-'], ['R']], 'th o x t e r'.split(),
local=True)
assert out[1] == '-' and out[-2] == '-'
def test_class2tokens(self):
classes = 'T-VKTV-R'
tokens = 'tʰ ɔ x t ə r'.split(' ')
out = class2tokens(classes, tokens)
_ = class2tokens([['T'], ['-VKTV-'], ['R']],
'th o x t e r'.split(),
local=True)
assert out[1] == '-' and out[-2] == '-'
def test_class2tokens():
classes = 'T-VKTV-R'
tokens = 'tʰ ɔ x t ə r'.split(' ')
out = class2tokens(classes, tokens)
assert out[1] == '-' and out[-2] == '-'
def test_class2tokens():
classes = 'T-VKTV-R'
tokens = 'tʰ ɔ x t ə r'.split(' ')
out = class2tokens(classes, tokens)
assert out[1] == '-' and out[-2] == '-'
def get_correspondences(alms, ref='lexstatid'):
"""
Compute sound correspondences for a given set of aligned cognates.
"""
# store all correspondences
corrs = {}
# store occurrences
occs = {}
for key, msa in alms.msa[ref].items():
# get basic stuff
idxs = msa['ID']
taxa = msa['taxa']
concept = cgi.escape(alms[idxs[0], 'concept'], True)
# get numerical representation of alignments
if 'numbers' in alms.header:
alignment = [class2tokens(
alms[idxs[i], 'numbers'],
msa['alignment'][i]) for i in range(len(idxs))]
else:
alignment = msa['alignment']
# create new array for confidence
character_matrix = []
# iterate over each taxon
for i, taxon in enumerate(taxa):
# get the numerical sequence
nums = alignment[i]
# store chars per line
chars = []
# iterate over the sequence
for j, num in enumerate(nums):
col = [alm[j] for alm in alignment]
# get the char
if num != '-':
charA = dotjoin(taxa[i], msa['alignment'][i][j], num.split('.')[2])
chars += [charA]
try:
occs[charA] += [concept]
except:
occs[charA] = [concept]
else:
chars += ['-']
for k, numB in enumerate(col):
if k != i:
if num == '-' and numB == '-':
pass
else:
if numB != '-' and num != '-':
# get the second char
charB = dotjoin(
taxa[k],
msa['alignment'][k][j],
numB.split('.')[2])
try:
corrs[charA][charB] += 1
except:
try:
corrs[charA][charB] = 1
except:
corrs[charA] = {charB: 1}
character_matrix += [chars]
# append confidence matrix to alignments
alms.msa[ref][key]['_charmat'] = character_matrix
return corrs, occs
def get_confidence(alms, scorer, ref='lexstatid', gap_weight=1):
"""
Function creates confidence scores for a given set of alignments.
Parameters
----------
alms : :py:class`~lingpy.align.sca.Alignments`
An *Alignments* object containing already aligned strings.
scorer : :py:class:`~lingpy.algorithm._misc.ScoreDict`
A *ScoreDict* object which gives similarity scores for all segments in
the alignment.
ref : str (default="lexstatid")
The reference entry-type, referring to the cognate-set to be used for
the analysis.
"""
# store all values for average scores
values = []
# store all correspondences
corrs = {}
# store occurrences
occs = {}
for key, msa in alms.msa[ref].items():
# get basic stuff
idxs = msa['ID']
taxa = msa['taxa']
concept = cgi.escape(alms[idxs[0], 'concept'], True)
# get numerical representation of alignments
if scorer:
alignment = [class2tokens(
alms[idxs[i], 'numbers'],
msa['alignment'][i]) for i in range(len(idxs))]
else:
alignment = msa['alignment']
# create new array for confidence
confidence_matrix = []
character_matrix = []
# iterate over each taxon
for i, taxon in enumerate(taxa):
idx = alms.taxa.index(taxon) + 1
# get the numerical sequence
nums = alignment[i]
# store confidences per line
confidences = []
# store chars per line
chars = []
# iterate over the sequence
for j, num in enumerate(nums):
col = [alm[j] for alm in alignment]
score = 0
count = 0
# get the char
if num != '-':
charA = dotjoin(taxa[i], msa['alignment'][i][j], num.split('.')[2])
chars += [charA]
try:
occs[charA] += [concept]
except:
occs[charA] = [concept]
else:
chars += ['-']
for k, numB in enumerate(col):
if k != i:
if num == '-' and numB == '-':
pass
else:
if numB != '-' and num != '-':
# get the second char
charB = dotjoin(
taxa[k], msa['alignment'][k][j], numB.split('.')[2])
try:
corrs[charA][charB] += 1
except:
try:
corrs[charA][charB] = 1
except:
corrs[charA] = {charB: 1}
gaps = False
if num == '-' and numB != '-':
numA = charstring(idx)
gaps = True
elif numB == '-' and num != '-':
numB = charstring(alms.taxa.index(taxa[k]))
numA = num
gaps = True
else:
numA = num
scoreA = scorer[numA, numB]
scoreB = scorer[numB, numA]
this_score = max(scoreA, scoreB)
if not gaps:
score += this_score
count += 1
else:
score += this_score * gap_weight
count += gap_weight
if count:
score = score / count
else:
score = -25
confidences += [int(score + 0.5)]
values += [int(score + 0.5)]
confidence_matrix += [confidences]
character_matrix += [chars]
# append confidence matrix to alignments
alms.msa[ref][key]['confidence'] = confidence_matrix
alms.msa[ref][key]['_charmat'] = character_matrix
# sort the values
values = sorted(set(values + [1]))
# make conversion to scale of 100 values
converter = {}
valsA = values[:values.index(1)]
valsB = values[values.index(1):]
stepA = 50 / (len(valsA) + 1)
stepB = 75 / (len(valsB) + 1)
for i, score in enumerate(valsA): # values[:values.index(0)):
converter[score] = int((stepA * i) / 4 + 0.5)
for i, score in enumerate(valsB):
converter[score] = int(stepB * i + 0.5) + 50
# iterate over keys again
for key, msa in alms.msa[ref].items():
# get basic stuff
for i, line in enumerate(msa['confidence']):
for j, cell in enumerate(line):
alms.msa[ref][key]['confidence'][i][j] = converter[cell]
jsond = {}
for key, corr in corrs.items():
splits = [c.split('.') + [o] for c, o in corr.items()]
sorts = sorted(splits, key=lambda x: (x[0], -x[3]))
new_sorts = []
# check for rowspan
spans = {}
for a, b, c, d in sorts:
if a in spans:
if spans[a] < 3 and d > 1:
spans[a] += 1
new_sorts += [[a, b, c, d]]
else:
if d > 1:
spans[a] = 1
new_sorts += [[a, b, c, d]]
bestis = []
old_lang = ''
counter = 0
for a, b, c, d in new_sorts:
new_lang = a
if new_lang != old_lang:
old_lang = new_lang
tmp = '<tr class="display">'
tmp += '<td class="display" rowspan={0}>'.format(spans[a])
tmp += a + '</td>'
tmp += '<td class="display" onclick="show({0});"><span '.format(
"'" + dotjoin(a, b, c) + "'")
tmp += 'class="char {0}">' + b + '</span></td>'
tmp += '<td class="display">'
tmp += c + '</td>'
tmp += '<td class="display">' + str(d) + '</td>'
tmp += '<td class="display">' + str(len(occs[dotjoin(a, b, c)])) + '</td>'
tmp += '</tr>'
t = 'dolgo_' + token2class(b, rcParams['dolgo'])
# bad check for three classes named differently
if t == 'dolgo__':
t = 'dolgo_X'
elif t == 'dolgo_1':
t = 'dolgo_TONE'
elif t == 'dolgo_0':
t = 'dolgo_ERROR'
bestis += [tmp.format(t)]
counter += 1
elif counter > 0:
tmp = '<tr class="display">'
tmp += '<td class="display" onclick="show({0});"><span '.format(
"'" + dotjoin(a, b, c) + "'")
tmp += 'class="char {0}">' + b + '</span></td>'
tmp += '<td class="display">' + c + '</td>'
tmp += '<td class="display">' + str(d) + '</td>'
tmp += '<td class="display">' + str(len(occs[dotjoin(a, b, c)])) + '</td>'
tmp += '</tr>'
t = 'dolgo_' + token2class(b, rcParams['dolgo'])
# bad check for three classes named differently
if t == 'dolgo__':
t = 'dolgo_X'
elif t == 'dolgo_1':
t = 'dolgo_TONE'
elif t == 'dolgo_0':
t = 'dolgo_ERROR'
bestis += [tmp.format(t)]
counter += 1
old_lang = new_lang
else:
old_lang = new_lang
counter = 0
jsond[key] = [''.join(bestis), occs[key]]
return jsond
def align(self, **keywords):
"""
Align a pair of sequences or multiple sequence pairs.
Parameters
----------
gop : int (default=-1)
The gap opening penalty (GOP).
scale : float (default=0.5)
The gap extension penalty (GEP), calculated with help of a scaling
factor.
mode : {"global","local","overlap","dialign"}
The alignment mode, see :evobib:`List2012a` for details.
factor : float (default = 0.3)
The factor by which matches in identical prosodic position are
increased.
restricted_chars : str (default="T\_")
The restricted chars that function as an indicator of syllable or
morpheme breaks for secondary alignment, see :evobib:`List2012c`
for details.
distance : bool (default=False)
If set to *True*, return the distance instead of the similarity
score. Distance is calculated using the formula by
:evobib:`Downey2008`.
model : { None, ~lingpy.data.model.Model }
Specify the sound class model that shall be used for the analysis.
If no model is specified, the default model of :evobib:`List2012a`
will be used.
pprint : bool (default=False)
If set to *True*, the alignments are printed to the screen.
"""
setdefaults(
keywords,
gop=-1,
scale=0.5,
mode='global',
factor=0.3,
restricted_chars='T_',
distance=False,
model=rcParams['sca'],
pprint=False,
transform=rcParams['align_transform'])
if hasattr(self, 'model'):
if keywords['model'] != self.model:
self._set_model(**keywords)
else:
self._set_model(**keywords)
# create the alignments array
self._alignments = calign.align_pairs(
self.classes,
self.weights,
self.prostrings,
keywords['gop'],
keywords['scale'],
keywords['factor'],
self.scoredict,
keywords['mode'],
keywords['restricted_chars'],
distance=1 if keywords['distance'] else 0)
# switch back to alignments
self.alignments = []
for i, (almA, almB, sim) in enumerate(self._alignments):
self.alignments.append((
class2tokens(self.tokens[i][0], almA, local=keywords['mode'] == "local"),
class2tokens(self.tokens[i][1], almB, local=keywords['mode'] == "local"),
sim))
# print the alignments, if this is chosen
as_string(self, pprint=keywords['pprint'])
def get_correspondences(alms, ref='lexstatid'):
"""
Compute sound correspondences for a given set of aligned cognates.
"""
# store all correspondences
corrs = {}
# store occurrences
occs = {}
for key, msa in alms.msa[ref].items():
# get basic stuff
idxs = msa['ID']
taxa = msa['taxa']
concept = cgi.escape(alms[idxs[0], 'concept'], True)
# get numerical representation of alignments
if 'numbers' in alms.header:
alignment = [class2tokens(
alms[idxs[i], 'numbers'],
msa['alignment'][i]) for i in range(len(idxs))]
else:
alignment = msa['alignment']
# create new array for confidence
character_matrix = []
# iterate over each taxon
for i, taxon in enumerate(taxa):
# get the numerical sequence
nums = alignment[i]
# store chars per line
chars = []
# iterate over the sequence
for j, num in enumerate(nums):
col = [alm[j] for alm in alignment]
# get the char
if num != '-':
charA = dotjoin(taxa[i], msa['alignment'][i][j], num.split('.')[2])
chars += [charA]
try:
occs[charA] += [concept]
except:
occs[charA] = [concept]
else:
chars += ['-']
for k, numB in enumerate(col):
if k != i:
if num == '-' and numB == '-':
pass
else:
if numB != '-' and num != '-':
# get the second char
charB = dotjoin(
taxa[k],
msa['alignment'][k][j],
numB.split('.')[2])
try:
corrs[charA][charB] += 1
except:
try:
corrs[charA][charB] = 1
except:
corrs[charA] = {charB: 1}
character_matrix += [chars]
# append confidence matrix to alignments
alms.msa[ref][key]['_charmat'] = character_matrix
return corrs, occs
def get_confidence(alms, scorer, ref='lexstatid', gap_weight=1):
"""
Function creates confidence scores for a given set of alignments.
Parameters
----------
alms : :py:class`~lingpy.align.sca.Alignments`
An *Alignments* object containing already aligned strings.
scorer : :py:class:`~lingpy.algorithm._misc.ScoreDict`
A *ScoreDict* object which gives similarity scores for all segments in
the alignment.
ref : str (default="lexstatid")
The reference entry-type, referring to the cognate-set to be used for
the analysis.
"""
# store all values for average scores
values = []
# store all correspondences
corrs = {}
# store occurrences
occs = {}
for key, msa in alms.msa[ref].items():
# get basic stuff
idxs = msa['ID']
taxa = msa['taxa']
concept = cgi.escape(alms[idxs[0], 'concept'], True)
# get numerical representation of alignments
if scorer:
alignment = [class2tokens(
alms[idxs[i], 'numbers'],
msa['alignment'][i]) for i in range(len(idxs))]
else:
alignment = msa['alignment']
# create new array for confidence
confidence_matrix = []
character_matrix = []
# iterate over each taxon
for i, taxon in enumerate(taxa):
idx = alms.taxa.index(taxon) + 1
# get the numerical sequence
nums = alignment[i]
# store confidences per line
confidences = []
# store chars per line
chars = []
# iterate over the sequence
for j, num in enumerate(nums):
col = [alm[j] for alm in alignment]
score = 0
count = 0
# get the char
if num != '-':
charA = dotjoin(taxa[i], msa['alignment'][i][j], num.split('.')[2])
chars += [charA]
try:
occs[charA] += [concept]
except:
occs[charA] = [concept]
else:
chars += ['-']
for k, numB in enumerate(col):
if k != i:
if num == '-' and numB == '-':
pass
else:
if numB != '-' and num != '-':
# get the second char
charB = dotjoin(
taxa[k], msa['alignment'][k][j], numB.split('.')[2])
try:
corrs[charA][charB] += 1
except:
try:
corrs[charA][charB] = 1
except:
corrs[charA] = {charB: 1}
gaps = False
if num == '-' and numB != '-':
numA = charstring(idx)
gaps = True
elif numB == '-' and num != '-':
numB = charstring(alms.taxa.index(taxa[k]))
numA = num
gaps = True
else:
numA = num
scoreA = scorer[numA, numB]
scoreB = scorer[numB, numA]
this_score = max(scoreA, scoreB)
if not gaps:
score += this_score
count += 1
else:
score += this_score * gap_weight
count += gap_weight
if count:
score = score / count
else:
score = -25
confidences += [int(score + 0.5)]
values += [int(score + 0.5)]
confidence_matrix += [confidences]
character_matrix += [chars]
# append confidence matrix to alignments
alms.msa[ref][key]['confidence'] = confidence_matrix
alms.msa[ref][key]['_charmat'] = character_matrix
# sort the values
values = sorted(set(values + [1]))
# make conversion to scale of 100 values
converter = {}
valsA = values[:values.index(1)]
valsB = values[values.index(1):]
stepA = 50 / (len(valsA) + 1)
stepB = 75 / (len(valsB) + 1)
for i, score in enumerate(valsA): # values[:values.index(0)):
converter[score] = int((stepA * i) / 4 + 0.5)
for i, score in enumerate(valsB):
converter[score] = int(stepB * i + 0.5) + 50
# iterate over keys again
for key, msa in alms.msa[ref].items():
# get basic stuff
for i, line in enumerate(msa['confidence']):
for j, cell in enumerate(line):
alms.msa[ref][key]['confidence'][i][j] = converter[cell]
jsond = {}
for key, corr in corrs.items():
splits = [c.split('.') + [o] for c, o in corr.items()]
sorts = sorted(splits, key=lambda x: (x[0], -x[3]))
new_sorts = []
# check for rowspan
spans = {}
for a, b, c, d in sorts:
if a in spans:
if spans[a] < 3 and d > 1:
spans[a] += 1
new_sorts += [[a, b, c, d]]
else:
if d > 1:
spans[a] = 1
new_sorts += [[a, b, c, d]]
bestis = []
old_lang = ''
counter = 0
for a, b, c, d in new_sorts:
new_lang = a
if new_lang != old_lang:
old_lang = new_lang
tmp = '<tr class="display">'
tmp += '<td class="display" rowspan={0}>'.format(spans[a])
tmp += a + '</td>'
tmp += '<td class="display" onclick="show({0});"><span '.format(
"'" + dotjoin(a, b, c) + "'")
tmp += 'class="char {0}">' + b + '</span></td>'
tmp += '<td class="display">'
tmp += c + '</td>'
tmp += '<td class="display">' + str(d) + '</td>'
tmp += '<td class="display">' + str(len(occs[dotjoin(a, b, c)])) + '</td>'
tmp += '</tr>'
t = 'dolgo_' + token2class(b, rcParams['dolgo'])
# bad check for three classes named differently
if t == 'dolgo__':
t = 'dolgo_X'
elif t == 'dolgo_1':
t = 'dolgo_TONE'
elif t == 'dolgo_0':
t = 'dolgo_ERROR'
bestis += [tmp.format(t)]
counter += 1
elif counter > 0:
tmp = '<tr class="display">'
tmp += '<td class="display" onclick="show({0});"><span '.format(
"'" + dotjoin(a, b, c) + "'")
tmp += 'class="char {0}">' + b + '</span></td>'
tmp += '<td class="display">' + c + '</td>'
tmp += '<td class="display">' + str(d) + '</td>'
tmp += '<td class="display">' + str(len(occs[dotjoin(a, b, c)])) + '</td>'
tmp += '</tr>'
t = 'dolgo_' + token2class(b, rcParams['dolgo'])
# bad check for three classes named differently
if t == 'dolgo__':
t = 'dolgo_X'
elif t == 'dolgo_1':
t = 'dolgo_TONE'
elif t == 'dolgo_0':
t = 'dolgo_ERROR'
bestis += [tmp.format(t)]
counter += 1
old_lang = new_lang
else:
old_lang = new_lang
counter = 0
jsond[key] = [''.join(bestis), occs[key]]
return jsond
def get_confidence(alms, scorer, ref='lexstatid', gap_weight=1):
"""
Function creates confidence scores for a given set of alignments.
Parameters
----------
alms : :py:class`~lingpy.align.sca.Alignments`
An *Alignments* object containing already aligned strings.
scorer : :py:class:`~lingpy.algorithm._misc.ScoreDict`
A *ScoreDict* object which gives similarity scores for all segments in
the alignment.
ref : str (default="lexstatid")
The reference entry-type, referring to the cognate-set to be used for
the analysis.
"""
# store all values for average scores
values = []
for key, msa in alms.msa[ref].items():
# get basic stuff
idxs = msa['ID']
taxa = msa['taxa']
# get numerical representation of alignments
if scorer:
alignment = [class2tokens(
alms[idxs[i], 'numbers'],
msa['alignment'][i]) for i in range(len(idxs))]
else:
alignment = msa['alignment']
# create new array for confidence
confidence_matrix = []
# iterate over each taxon
for i, taxon in enumerate(taxa):
idx = alms.taxa.index(taxon) + 1
# get the numerical sequence
nums = alignment[i]
# store confidences per line
confidences = []
# iterate over the sequence
for j, num in enumerate(nums):
col = [alm[j] for alm in alignment]
score = 0
count = 0
for k, numB in enumerate(col):
if k != i:
if num == '-' and numB == '-':
pass
else:
gaps = False
if num == '-' and numB != '-':
numA = str(idx) + '.X.-'
gaps = True
elif numB == '-' and num != '-':
numB = str(alms.taxa.index(taxa[k])) + '.X.-'
numA = num
gaps = True
else:
numA = num
scoreA = scorer[numA, numB]
scoreB = scorer[numB, numA]
this_score = max(scoreA, scoreB)
if not gaps:
score += this_score
count += 1
else:
score += this_score * gap_weight
count += gap_weight
if count:
score = score / count # (len(col) - gaps * gap_weight)
else:
score = -25
confidences += [int(score + 0.5)]
values += [int(score + 0.5)]
confidence_matrix += [confidences]
# append confidence matrix to alignments
alms.msa[ref][key]['confidence'] = confidence_matrix
# sort the values
values = sorted(set(values))
# make conversion to scale of 100 values
converter = {}
step = 100 / (len(values) + 1)
for i, score in enumerate(values):
converter[score] = int(step * score + 0.5)
# iterate over keys again
for key, msa in alms.msa[ref].items():
# get basic stuff
for i, line in enumerate(msa['confidence']):
for j, cell in enumerate(line):
msa['confidence'] = converter[cell]
def align(self, **keywords):
"""
Align a pair of sequences or multiple sequence pairs.
Parameters
----------
gop : int (default=-1)
The gap opening penalty (GOP).
scale : float (default=0.5)
The gap extension penalty (GEP), calculated with help of a scaling
factor.
mode : {"global","local","overlap","dialign"}
The alignment mode, see :evobib:`List2012a` for details.
factor : float (default = 0.3)
The factor by which matches in identical prosodic position are
increased.
restricted_chars : str (default="T_")
The restricted chars that function as an indicator of syllable or
morpheme breaks for secondary alignment, see :evobib:`List2012c`
for details.
distance : bool (default=False)
If set to *True*, return the distance instead of the similarity
score. Distance is calculated using the formula by
:evobib:`Downey2008`.
model : { None, ~lingpy.data.model.Model }
Specify the sound class model that shall be used for the analysis.
If no model is specified, the default model of :evobib:`List2012a`
will be used.
pprint : bool (default=False)
If set to *True*, the alignments are printed to the screen.
"""
setdefaults(
keywords,
gop=-1,
scale=0.5,
mode='global',
factor=0.3,
restricted_chars='T_',
distance=False,
model=rcParams['sca'],
pprint=False,
transform=rcParams['align_transform'])
if hasattr(self, 'model'):
if keywords['model'] != self.model:
self._set_model(**keywords)
else:
self._set_model(**keywords)
# create the alignments array
self._alignments = calign.align_pairs(
self.classes,
self.weights,
self.prostrings,
keywords['gop'],
keywords['scale'],
keywords['factor'],
self.scoredict,
keywords['mode'],
keywords['restricted_chars'],
distance=1 if keywords['distance'] else 0)
# switch back to alignments
self.alignments = []
for i, (almA, almB, sim) in enumerate(self._alignments):
self.alignments.append((
class2tokens(self.tokens[i][0], almA, local=keywords['mode'] == "local"),
class2tokens(self.tokens[i][1], almB, local=keywords['mode'] == "local"),
sim))
# print the alignments, if this is chosen
as_string(self, pprint=keywords['pprint'])