/usr/share/pyshared/frog/csidp.py is in frog 0.12.17-7.1+b1.
This file is owned by root:root, with mode 0o644.
The actual contents of the file can be viewed below.
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 | #!/usr/bin/env python
"""
Constraint satisfaction inference for dependency parsing.
usage: %prog [options] [file...]
-mDIST, --max-dist=DIST: maximum distance between head and dependent
-x, --exclude-non-scoring: do not generate instance for non-scoring
tokens
Constraints:
--dep=FILENAME: classifier output for the C_dep constraint
--dir=FILENAME: classifier output for the C_dir constraint
--mod=FILENAME: classifier output for the C_mod constraint
--out=FILENAME: output file name
Parsing algorithm:
--non-projective: approximate non-projective parsing
---
:Refinements:
-m: type='int', dest='maxDist'
-x: action='store_true', default=False, dest='skipNonScoring'
"""
import fileinput
import sys
from itertools import izip, imap, groupby, ifilter
from operator import itemgetter, attrgetter
from sentences import sentenceIterator
import cky
import csiparse2 as csiparse
import deptree
from common import *
def leftIncomplete(chart, s, t, sentence):
r = chart[s, t, "l", False].r
label = chart[s, t, "l", False].edgeLabel
if r is not None:
assert s > 0
sentence[s - 1][DEPREL] = label
sentence[s - 1][HEAD] = t
rightComplete(chart, s, r, sentence)
leftComplete(chart, r + 1, t, sentence)
def rightIncomplete(chart, s, t, sentence):
r = chart[s, t, "r", False].r
label = chart[s, t, "r", False].edgeLabel
if r is not None:
assert t > 0
sentence[t - 1][DEPREL] = label
sentence[t - 1][HEAD] = s
rightComplete(chart, s, r, sentence)
leftComplete(chart, r + 1, t, sentence)
def leftComplete(chart, s, t, sentence):
r = chart[s, t, "l", True].r
if r is not None:
leftComplete(chart, s, r, sentence)
leftIncomplete(chart, r, t, sentence)
def rightComplete(chart, s, t, sentence):
r = chart[s, t, "r", True].r
if r is not None:
rightIncomplete(chart, s, r, sentence)
rightComplete(chart, r, t, sentence)
def cyclic(sentence, head, dependent):
if head == dependent:
return True
x = sentence[head - 1][HEAD]
while x > 0:
if x == dependent:
return True
else:
x = sentence[x - 1][HEAD]
return False
def evaluateTree(sentence, parser):
score = 0
for i in xrange(len(sentence)):
inDeps = set(x[DEPREL] for x in sentence if x[HEAD] == i + 1)
outDep = sentence[i][DEPREL]
# C_mod
for constraint in parser.inDepConstraints[i + 1]:
if constraint.relType in inDeps:
score += constraint.weight
# C_dir
for constraint in parser.outDepConstraints[i + 1]:
if (0 < sentence[i][HEAD] < i + 1 and \
constraint.direction == constraint.LEFT) or \
(sentence[i][HEAD] > i + 1 and \
constraint.direction == constraint.RIGHT) or \
(sentence[i][HEAD] == 0 and \
constraint.direction == constraint.ROOT):
score += constraint.weight
# C_dep
if True: #sentence[i][HEAD] > 0:
for constraint in parser.edgeConstraints[i + 1][sentence[i][HEAD]]:
if constraint.relType == sentence[i][DEPREL]:
score += constraint.weight
return score
def scoreDiff(sentence, parser, dependent, newHead):
result = 0
oldHead = sentence[dependent - 1][HEAD]
oldRel = sentence[dependent - 1][DEPREL]
# C_dep diff
if True: #oldHead >= 0:
depRels = parser.edgeConstraints[dependent][oldHead]
assert len(depRels) <= 1
if depRels and depRels[0].relType == oldRel:
result -= depRels[0].weight
if True: #newHead >= 0:
depRels = parser.edgeConstraints[dependent][newHead]
assert len(depRels) <= 1
if depRels:
newRel = depRels[0].relType
result += depRels[0].weight
else:
newRel = "unk" if newHead > 0 else "ROOT"
# C_mod diff
if oldHead > 0:
dependents = [x
for x in sentence
if x[HEAD] == oldHead and x[DEPREL] == oldRel]
if len(dependents) == 1:
for constraint in parser.inDepConstraints[oldHead]:
if constraint.relType == oldRel:
result -= constraint.weight
if newHead > 0:
dependents = [x
for x in sentence
if x[HEAD] == newHead and x[DEPREL] == newRel]
if len(dependents) == 0:
for constraint in parser.inDepConstraints[newHead]:
if constraint.relType == newRel:
result += constraint.weight
# C_dir diff
if True: #(dependent - oldHead) * (dependent - newHead) < 0:
for constraint in parser.outDepConstraints[dependent]:
if (0 < oldHead < dependent and \
constraint.direction == constraint.LEFT) or \
(oldHead > dependent and \
constraint.direction == constraint.RIGHT) or \
(oldHead == 0 and \
constraint.direction == constraint.ROOT):
result -= constraint.weight
if (0 < newHead < dependent and \
constraint.direction == constraint.LEFT) or \
(newHead > dependent and \
constraint.direction == constraint.RIGHT) or \
(newHead == 0 and \
constraint.direction == constraint.ROOT):
result += constraint.weight
return result
def approxNonProjective(sentence, parser):
#currentScore = 0 #evaluateTree(sentence, parser)
currentScore = evaluateTree(sentence, parser)
while True:
bestScore = float("-inf")
bestHead = None
bestDependent = None
for j in xrange(1, len(sentence) + 1):
for i in xrange(0, len(sentence) + 1):
assert type(sentence[j - 1][HEAD]) == int
if not sentence[j - 1][HEAD] == i and \
not cyclic(sentence, i, j):
score = currentScore + scoreDiff(sentence, parser, j, i)
########
#newSentence = [t[:] for t in sentence]
#newSentence[j - 1][HEAD] = i
#if True: #i > 0:
# depRels = parser.edgeConstraints[j][i]
# assert len(depRels) <= 1
# if depRels:
# newSentence[j - 1][DEPREL] = depRels[0].relType
# else:
# newSentence[j - 1][DEPREL] = "unk" if i > 0 else "ROOT"
#assert abs(evaluateTree(newSentence, parser) - score) < 0.0001, ("%s %s" % (j, i))
#######
if score > bestScore:
bestScore = score
bestHead = i
bestDependent = j
if bestScore > currentScore:
# sys.stderr.write("+")
currentScore = bestScore
sentence[bestDependent - 1][HEAD] = bestHead
if True: #bestHead > 0:
depRels = parser.edgeConstraints[bestDependent][bestHead]
assert len(depRels) <= 1
if depRels:
sentence[bestDependent - 1][DEPREL] = depRels[0].relType
else:
sentence[bestDependent - 1][DEPREL] = "unk" if bestHead > 0 else "ROOT"
#else:
# sentence[bestDependent - 1][DEPREL] = "ROOT"
else:
return
def main(args):
import cmdline
main_(*cmdline.parse(__doc__, args))
def main_(options, args):
assert options.dep
assert options.out
pairsStream = open(options.dep)
pairsIterator = csiparse.instanceIterator(pairsStream)
# print >> sys.stderr, "C_dep constraints enabled"
if options.dir:
dirStream = open(options.dir)
dirIterator = csiparse.instanceIterator(dirStream)
# print >> sys.stderr, "C_dir constraints enabled"
else:
dirIterator = None
if options.mod:
relsStream = open(options.mod)
relsIterator = csiparse.instanceIterator(relsStream)
# print >> sys.stderr, "C_mod constraints enabled"
else:
relsIterator = None
for sentence in sentenceIterator(fileinput.input(args)):
outfile = open(options.out, "w" )
domains, constraints = csiparse.formulateWCSP(sentence,
dirIterator,
relsIterator,
pairsIterator,
options)
parser = cky.CKYParser(len(sentence))
for constraint in constraints:
parser.addConstraint(constraint)
chart = parser.parse()
#item = chart[0, self.numTokens - 1, "r", True]
#for token in sentence:
# token[DEPREL] = "__"
# token[HEAD] = "__"
for token in sentence:
if len(token) <= DEPREL:
token.extend([None, None])
#print chart[0, len(sentence) - 1, "r", True].r + 1
rightComplete(chart, 0, len(sentence) - 1 + 1, sentence)
if options.non_projective:
approxNonProjective(sentence, parser)
for token in sentence:
outfile.write( " ".join(map(str, token)) )
outfile.write("\n")
# sys.stderr.write(".")
if __name__ == "__main__":
import cmdline
main(*cmdline.parse())
|