1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
|
#!/usr/bin/env python
"""Leaf and Edge classes that can calculate their likelihoods.
Each leaf holds a sequence. Used by a likelihood function."""
from cogent.util.modules import importVersionedModule, ExpectedImportError
from cogent.util.parallel import MPI
import numpy
numpy.seterr(all='ignore')
numerictypes = numpy.core.numerictypes.sctype2char
__author__ = "Peter Maxwell"
__copyright__ = "Copyright 2007-2009, The Cogent Project"
__credits__ = ["Peter Maxwell", "Rob Knight"]
__license__ = "GPL"
__version__ = "1.5.0"
__maintainer__ = "Peter Maxwell"
__email__ = "pm67nz@gmail.com"
__status__ = "Production"
try:
pyrex = importVersionedModule('_likelihood_tree', globals(),
(2, 1), "pure Python/NumPy likelihood tree")
except ExpectedImportError:
pyrex = None
class _LikelihoodTreeEdge(object):
def __init__(self, children, edge_name, alignment=None):
self.edge_name = edge_name
self.alphabet = children[0].alphabet
self.comm = None # for MPI
M = children[0].shape[-1]
for child in children:
assert child.shape[-1] == M
# Unique positions are unique combos of input positions
if alignment is None:
# The children are pre-aligned gapped sequences
assignments = [c.index for c in children]
else:
self.alignment = alignment #XXX preserve through MPI split?
# The children are ungapped sequences, 'alignment'
# indicates where gaps need to go.
assignments = []
for (i, c) in enumerate(children):
a = []
for align_index in alignment:
col = align_index[i]
if col is None:
u = len(c.uniq)-1 # gap
else:
u = c.index[col]
assert 0 <= u < len(c.uniq)-1, (
u, len(c.uniq), c.uniq[-1], align_index)
a.append(u)
assignments.append(a)
(uniq, counts, self.index) = _indexed(zip(*assignments))
# extra column for gap
uniq.append(tuple([len(c.uniq)-1 for c in children]))
counts.append(0)
self.uniq = numpy.asarray(uniq, self.integer_type)
# For faster math, a contiguous index array for each child
self.indexes = [
numpy.array(list(ch), self.integer_type)
for ch in numpy.transpose(self.uniq)]
# If this is the root it will need to weight the total
# log likelihoods by these counts:
self.counts = numpy.array(counts, self.float_type)
# For product of child likelihoods
self._indexed_children = zip(self.indexes, children)
self.shape = [len(self.uniq), M]
def restrictMotif(self, input_likelihoods, fixed_motif):
# for reconstructAncestralSeqs
mask = numpy.zeros([input_likelihoods.shape[-1]], self.float_type)
mask[fixed_motif] = 1.0
input_likelihoods *= mask
def parallelShare(self, comm):
"""A local version of self for a single CPU in an MPI group"""
if comm is None or comm.Get_size() == 1:
return self
assert self.comm is None
U = len(self.uniq) - 1 # Gap column
(size, rank) = (comm.Get_size(), comm.Get_rank())
(share, remainder) = divmod(U, size)
if share == 0:
return self # not enough to share
share_sizes = [share+1]*remainder + [share]*(size-remainder)
assert sum(share_sizes) == U
(lo,hi) = [sum(share_sizes[:i]) for i in (rank, rank+1)]
local_cols = [i for (i,u) in enumerate(self.index)
if lo <= u < hi]
local = self.selectColumns(local_cols)
# Attributes for reconstructing the global array.
# should probably make a wrapping class instead.
local.share_sizes = share_sizes
local.index = self.index # yuk
local.comm = comm
return local
def selectColumns(self, cols):
children = []
for (index, child) in self._indexed_children:
child = child.selectColumns(cols)
children.append(child)
return self.__class__(children, self.edge_name)
def parallelReconstructColumns(self, llh):
"""Recombine full uniq array (eg: likelihoods) from MPI CPUs"""
if self.comm is None:
return llh
result = numpy.empty([sum(self.share_sizes)], llh.dtype)
send = (llh[:-1], MPI.DOUBLE) # drop gap column
recv = (result, self.share_sizes, None, MPI.DOUBLE)
self.comm.Allgatherv(send, recv)
return result
def getFullLengthLikelihoods(self, input_likelihoods):
lh = self.parallelReconstructColumns(input_likelihoods)
return numpy.take(lh, self.index, 0)
def getEdge(self, name):
if self.edge_name == name:
return self
else:
for (i,c) in self._indexed_children:
r = c.getEdge(name)
if r is not None:
return r
return None
def makePartialLikelihoodsArray(self):
return numpy.ones(self.shape, self.float_type)
def sumInputLikelihoods(self, *likelihoods):
result = numpy.ones(self.shape, self.float_type)
self.sumInputLikelihoodsR(result, *likelihoods)
return result
def asLeaf(self, likelihoods):
assert len(likelihoods) == len(self.counts)
return LikelihoodTreeLeaf(likelihoods, likelihoods,
self.counts, self.index, self.edge_name, self.alphabet, None)
class _PyLikelihoodTreeEdge(_LikelihoodTreeEdge):
# Should be a subclass of regular tree edge?
float_type = numerictypes(float)
integer_type = numerictypes(int)
# For scaling very very small numbers
BASE = 2.0 ** 100
LOG_BASE = numpy.log(BASE)
def sumInputLikelihoodsR(self, result, *likelihoods):
result[:] = 1.0
for (i, index) in enumerate(self.indexes):
result *= numpy.take(likelihoods[i], index, 0)
return result
def logDotReduce(self, patch_probs, switch_probs, plhs):
plhs = self.parallelReconstructColumns(plhs)
exponent = 0
state_probs = patch_probs.copy()
for site in self.index:
state_probs = numpy.dot(switch_probs, state_probs) * plhs[site]
while max(state_probs) < 1.0:
state_probs *= self.BASE
exponent -= 1
return numpy.log(sum(state_probs)) + exponent * self.LOG_BASE
def getTotalLogLikelihood(self, input_likelihoods, mprobs):
lhs = numpy.inner(input_likelihoods, mprobs)
return self.getLogSumAcrossSites(lhs)
def getLogSumAcrossSites(self, lhs):
#print 'lhs, log(lhs)', lhs, numpy.log(lhs)
return numpy.inner(numpy.log(lhs), self.counts)
class _PyxLikelihoodTreeEdge(_LikelihoodTreeEdge):
integer_type = numerictypes(int) # match checkArrayInt1D
float_type = numerictypes(float) # match checkArrayDouble1D/2D
def sumInputLikelihoodsR(self, result, *likelihoods):
pyrex.sumInputLikelihoods(self.indexes, result, likelihoods)
return result
# For root
def logDotReduce(self, patch_probs, switch_probs, plhs):
plhs = self.parallelReconstructColumns(plhs)
return pyrex.logDotReduce(self.index, patch_probs, switch_probs, plhs)
def getTotalLogLikelihood(self, input_likelihoods, mprobs):
return pyrex.getTotalLogLikelihood(self.counts, input_likelihoods,
mprobs)
def getLogSumAcrossSites(self, lhs):
return pyrex.getLogSumAcrossSites(self.counts, lhs)
if pyrex is None:
LikelihoodTreeEdge = _PyLikelihoodTreeEdge
else:
LikelihoodTreeEdge = _PyxLikelihoodTreeEdge
FLOAT_TYPE = LikelihoodTreeEdge.float_type
INTEGER_TYPE = LikelihoodTreeEdge.integer_type
def _indexed(values):
# >>> _indexed(['a', 'b', 'c', 'a', 'a'])
# (['a', 'b', 'c'], [3, 1, 1], [0, 1, 2, 0, 0])
index = numpy.zeros([len(values)], INTEGER_TYPE)
unique = []
counts = []
seen = {}
for (c, key) in enumerate(values):
if key in seen:
i = seen[key]
counts[i] += 1
else:
i = len(unique)
unique.append(key)
counts.append(1)
seen[key] = i
index[c] = i
return unique, counts, index
def makeLikelihoodTreeLeaf(sequence, alphabet=None, seq_name=None):
if alphabet is None:
alphabet = sequence.MolType.Alphabet
if seq_name is None:
seq_name = sequence.getName()
motif_len = alphabet.getMotifLen()
sequence2 = sequence.getInMotifSize(motif_len)
# Convert sequence to indexed list of unique motifs
(uniq_motifs, counts, index) = _indexed(sequence2)
# extra column for gap
uniq_motifs.append('?' * motif_len)
counts.append(0)
counts = numpy.array(counts, FLOAT_TYPE)
# Convert list of unique motifs to array of unique profiles
try:
likelihoods = alphabet.fromAmbigToLikelihoods(
uniq_motifs, FLOAT_TYPE)
except alphabet.AlphabetError, detail:
motif = str(detail)
posn = list(sequence2).index(motif) * motif_len
raise ValueError, '%s at %s:%s not in alphabet' % (
repr(motif), seq_name, posn)
return LikelihoodTreeLeaf(uniq_motifs, likelihoods,
counts, index, seq_name, alphabet, sequence)
class LikelihoodTreeLeaf(object):
def __init__(self, uniq, likelihoods, counts, index, edge_name,
alphabet, sequence):
if sequence is not None:
self.sequence = sequence
self.alphabet = alphabet
self.name = self.edge_name = edge_name
self.uniq = uniq
self.input_likelihoods = likelihoods
self.counts = counts
self.index = index
self.shape = likelihoods.shape
self.ambig = numpy.sum(self.input_likelihoods, axis=-1)
# uniq (list of motifs) is redundant in that it could be derived
# from input_likelihoods (array of profiles) but it's easier just
# keep it rather than regenerate it. Also it's only used for
# self.getAmbiguousPositions(), so it might be nice to get rid of
# it eventually.
def backward(self):
index = numpy.array(self.index[::-1,...])
result = self.__class__(self.uniq, self.input_likelihoods, self.counts,
index, self.edge_name, self.alphabet, None)
return result
def __len__(self):
return len(self.index)
def __getitem__(self, index):
cols = range(*index.indices(len(self.index)))
return self.selectColumns(cols)
def getMotifCounts(self, include_ambiguity=False):
weights = self.counts / self.ambig
profile = self.input_likelihoods * weights[...,numpy.newaxis]
if not include_ambiguity:
unambig = self.ambig == 1.0
profile = numpy.compress(unambig, profile, axis=0)
return numpy.sum(profile, axis=0)
def getAmbiguousPositions(self):
ambig = {}
for (i,u) in enumerate(self.index):
if self.ambig[u] != 1.0:
ambig[i] = self.uniq[u]
return ambig
def selectColumns(self, cols):
sub_index = [self.index[i] for i in cols]
(keep, counts, index) = _indexed(sub_index)
keep.append(len(self.uniq)-1) # extra column for gap
counts.append(0)
counts = numpy.array(counts, FLOAT_TYPE)
uniq = [self.uniq[u] for u in keep]
likelihoods = numpy.take(self.input_likelihoods, keep, 0)
return self.__class__(
uniq, likelihoods, counts, index, self.edge_name,
self.alphabet, None)
def getEdge(self, name):
if self.edge_name == name:
return self
else:
return None
|