This repository has been archived by the owner on Feb 19, 2020. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 82
/
EnumeratedAnchoring.scala
275 lines (244 loc) · 10 KB
/
EnumeratedAnchoring.scala
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
package epic.parser
package projections
/*
Copyright 2012 David Hall
Licensed under the Apache License, Version 2.0 (the "License")
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
*/
import java.io._
import projections.AnchoredForestProjector.ForestData
import breeze.collection.mutable.{OpenAddressHashArray, TriangularArray}
import epic.lexicon.Lexicon
import epic.constraints.ChartConstraints
import epic.trees.Span
import java.util
/**
* Creates a locally-normalized anchored PCFG from some refined forest.
* @author dlwh
*/
case class AnchoredPCFGProjector[L, W](threshold: Double = Double.NegativeInfinity) extends ChartProjector[L, W] {
type MyAnchoring = EnumeratedAnchoring[L, W]
private def normalize(grammar: RuleTopology[L], ruleScores: OpenAddressHashArray[Double], totals: OpenAddressHashArray[Double]):OpenAddressHashArray[Double] = {
if (ruleScores eq null) null
else {
val r = new OpenAddressHashArray[Double](ruleScores.length, Double.NegativeInfinity, ruleScores.activeSize)
for( (rule, score) <- ruleScores.activeIterator) {
val parent = grammar.parent(rule)
if (score > 0)
r(rule) = math.log(score) - math.log(totals(parent))
}
r
}
}
private def logify(ruleScores: OpenAddressHashArray[Double]):OpenAddressHashArray[Double] = {
if (ruleScores eq null) null
else {
val r = new OpenAddressHashArray[Double](ruleScores.length, Double.NegativeInfinity, ruleScores.activeSize)
for( (rule, score) <- ruleScores.activeIterator) {
r(rule) = math.log(score)
}
r
}
}
protected def createAnchoring(charts: ParseMarginal[L, W], ruleData: ForestData, sentProb: Double) = {
val AnchoredForestProjector.ForestData(lexicalScores, unaryScores, totalsUnaries, binaryScores, totalsBinaries) = ruleData
val normUnaries:Array[OpenAddressHashArray[Double]] = for((ruleScores, totals) <- unaryScores zip totalsUnaries) yield {
normalize(charts.topology, ruleScores, totals)
}
val normBinaries:Array[Array[OpenAddressHashArray[Double]]] = for ((splits, totals) <- binaryScores zip totalsBinaries) yield {
if (splits eq null) null
else for(ruleScores <- splits) yield normalize(charts.topology, ruleScores, totals)
}
val sparsity = charts.anchoring.sparsityPattern
new EnumeratedAnchoring(charts.topology, charts.lexicon, charts.words, lexicalScores.map(logify), normUnaries, normBinaries, sparsity)
}
}
/**
* Creates anchorings for a set of trees from some parser using p(rule | sentence) marginals.
* @author dlwh
*/
@SerialVersionUID(469174684243960202L)
case class AnchoredRuleMarginalProjector[L, W](threshold: Double = Double.NegativeInfinity) extends ChartProjector[L, W] {
private def normalize(ruleScores: OpenAddressHashArray[Double]):OpenAddressHashArray[Double] = {
if (ruleScores eq null) null
else {
val r = new OpenAddressHashArray[Double](ruleScores.length, Double.NegativeInfinity, ruleScores.activeSize)
for( (rule, score) <- ruleScores.activeIterator) {
r(rule) = math.log(score)
}
r
}
}
type MyAnchoring = EnumeratedAnchoring[L, W]
protected def createAnchoring(charts: ParseMarginal[L, W],
ruleData: ForestData,
sentProb: Double) = {
val AnchoredForestProjector.ForestData(lexicalScores, unaryScores, _, binaryScores, _) = ruleData
val normUnaries:Array[OpenAddressHashArray[Double]] = unaryScores.map(normalize)
val normBinaries:Array[Array[OpenAddressHashArray[Double]]] = for (splits <- binaryScores) yield {
if (splits eq null) null
else splits.map(normalize)
}
val sparsity = charts.anchoring.sparsityPattern
new EnumeratedAnchoring(charts.topology, charts.lexicon, charts.words, lexicalScores.map(normalize), normUnaries, normBinaries, sparsity)
}
}
/**
* An [[epic.parser.projections.EnumeratedAnchoring]] is a compressed forest, with potentially distinct
* rule and span scores for every possible anchoring into the sentence. Commonly produced by either
* an [[epic.parser.projections.AnchoredPCFGProjector]] (for PCFG forests) or an [[epic.parser.projections.AnchoredRuleMarginalProjector]]
* (for rule marginal forests).
* @tparam L
* @tparam W
*/
@SerialVersionUID(3)
case class EnumeratedAnchoring[L, W](topology: RuleTopology[L],
lexicon: Lexicon[L, W],
words: IndexedSeq[W],
spanScores: Array[OpenAddressHashArray[Double]], // triangular index -> label -> score
// (begin, end) -> rule -> score
unaryScores: Array[OpenAddressHashArray[Double]],
// (begin, end) -> (split-begin) -> rule -> score
binaryScores: Array[Array[OpenAddressHashArray[Double]]],
sparsityPattern: ChartConstraints[L]) extends UnrefinedGrammarAnchoring[L, W] with Serializable {
override def addConstraints(cs: ChartConstraints[L]): UnrefinedGrammarAnchoring[L, W] = copy(sparsityPattern = sparsityPattern & cs)
/**
* Computes the pointwise division of two grammars, augmenting
* their refinement space to reflect this. If they share the same annotationTag,
* (assuming it's non-negative) they will share their state space. (That is,
* they will have the same annotations.)
* @param other
* @return
*/
override def /(other: GrammarAnchoring[L, W]): GrammarAnchoring[L, W] = {
other match {
case that: EnumeratedAnchoring[L, W] => EnumeratedAnchoring.divide(this, that)
case _ => super./(other)
}
}
/**
* Computes the point-wise division of this grammar with some other grammar.
*
* Note that scores are in log space, so we actually subtract scores.
* @param other
* @return
*/
override def /(other: UnrefinedGrammarAnchoring[L, W]): UnrefinedGrammarAnchoring[L, W] = {
other match {
case that: EnumeratedAnchoring[L, W] => EnumeratedAnchoring.divide(this, that)
case _ => super./(other)
}
}
def scoreUnaryRule(begin: Int, end: Int, rule: Int) = {
val forSpan = unaryScores(TriangularArray.index(begin, end))
if (forSpan eq null) Double.NegativeInfinity
else forSpan(rule)
}
def scoreBinaryRule(begin: Int, split: Int, end: Int, rule: Int) = {
val ti = TriangularArray.index(begin, end)
val forSpan = binaryScores(ti)
val cached = checkCache(split, rule, ti)
if (!java.lang.Double.isNaN(cached)) {
cached
} else if (forSpan eq null) {
Double.NegativeInfinity
} else {
val forSplit = forSpan(split - begin)
val result = if (forSplit eq null) Double.NegativeInfinity
else forSplit(rule)
updateCache(split, rule, ti, result)
result
}
}
def scoreSpan(begin: Int, end: Int, tag: Int): Double = {
val scores = spanScores(TriangularArray.index(begin, end))
if (scores ne null) scores(tag)
else Double.NegativeInfinity
}
// (1 entry for each position (a split point), an entry has a rule index, a begin/end pair, and a score
private val cache = new Array[Int](length * (1 + 1 + 2))
util.Arrays.fill(cache, -1) //
private def checkCache(splitPoint: Int, rule: Int, ti: Int) = {
val crule = cache(splitPoint * 4)
val cti = cache(splitPoint * 4 + 1)
if (rule == crule && cti == ti) {
java.lang.Double.longBitsToDouble(Span(cache(splitPoint * 4 + 2), cache(splitPoint * 4 + 3)).encoded)
} else {
Double.NaN
}
}
private def updateCache(splitPoint: Int, rule: Int, ti: Int, score: Double):Unit = {
cache(splitPoint * 4) = rule
cache(splitPoint * 4 + 1) = ti
val asSpan = new Span(java.lang.Double.doubleToRawLongBits(score))
cache(splitPoint * 4 + 2) = asSpan.begin
cache(splitPoint * 4 + 3) = asSpan.end
}
}
object EnumeratedAnchoring {
def divide[L, W](a: EnumeratedAnchoring[L, W], b: EnumeratedAnchoring[L, W]): EnumeratedAnchoring[L, W] = {
val newSpanScores = Array.tabulate(a.spanScores.length) { i =>
val oldA = a.spanScores(i)
val oldB = b.spanScores(i)
if (null == oldA || null == oldB) {
null
} else {
doDivide(oldA, oldB)
}
}
val newUnaryScores = Array.tabulate(a.unaryScores.length) { i =>
val oldA = a.unaryScores(i)
val oldB = b.unaryScores(i)
if (null == oldA || null == oldB) {
null
} else {
doDivide(oldA, oldB)
}
}
val newBinaryScores = Array.tabulate(a.binaryScores.length) { i =>
val aArray = a.binaryScores(i)
val bArray = b.binaryScores(i)
if (null == aArray || null == bArray) {
null
} else {
Array.tabulate(aArray.length) { split =>
val oldA = aArray(split)
val oldB = bArray(split)
if (null == oldA || null == oldB) {
null
} else {
doDivide(oldA, oldB)
}
}
}
}
a.copy(spanScores = newSpanScores, unaryScores=newUnaryScores, binaryScores = newBinaryScores)
}
private def doDivide(a: OpenAddressHashArray[Double], b: OpenAddressHashArray[Double]) = {
if (a == null || b == null) {
null
} else {
val oah = new OpenAddressHashArray[Double](a.size, a.default, a.activeSize min b.activeSize)
var off = 0
while (off < a.iterableSize) {
if (a.isActive(off)) {
val aa = a.valueAt(off)
val ii = a.indexAt(off)
val bb = b(ii)
if (aa != Double.NegativeInfinity && bb != Double.NegativeInfinity) {
oah(ii) = aa - bb
}
}
off += 1
}
oah
}
}
}