summaryrefslogtreecommitdiff
path: root/python/cdec/hypergraph.pxi
blob: 16a68a00685a295659e601fce95f514526833763 (plain)
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
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
cimport hypergraph
cimport kbest

cdef class Hypergraph:
    cdef hypergraph.Hypergraph* hg
    cdef MT19937* rng

    def __dealloc__(self):
        del self.hg
        if self.rng != NULL:
            del self.rng

    cdef MT19937* _rng(self):
        if self.rng == NULL:
            self.rng = new MT19937()
        return self.rng

    def viterbi(self):
        """hg.viterbi() -> String for the best hypothesis in the hypergraph."""
        cdef vector[WordID] trans
        hypergraph.ViterbiESentence(self.hg[0], &trans)
        return unicode(GetString(trans).c_str(), 'utf8')

    def viterbi_trees(self):
        """hg.viterbi_trees() -> (f_tree, e_tree)
        f_tree: Source tree for the best hypothesis in the hypergraph.
        e_tree: Target tree for the best hypothesis in the hypergraph.
        """
        f_tree = unicode(hypergraph.ViterbiFTree(self.hg[0]).c_str(), 'utf8')
        e_tree = unicode(hypergraph.ViterbiETree(self.hg[0]).c_str(), 'utf8')
        return (f_tree, e_tree)
    
    def viterbi_features(self):
        """hg.viterbi_features() -> SparseVector with the features corresponding
        to the best derivation in the hypergraph."""
        cdef SparseVector fmap = SparseVector.__new__(SparseVector)
        fmap.vector = new FastSparseVector[weight_t](hypergraph.ViterbiFeatures(self.hg[0]))
        return fmap

    def viterbi_forest(self):
        cdef Hypergraph hg = Hypergraph()
        hg.hg = new hypergraph.Hypergraph(self.hg[0].CreateViterbiHypergraph(NULL).get()[0])
        return hg

    def viterbi_joshua(self):
        """hg.viterbi_joshua() -> Joshua representation of the best derivation."""
        return unicode(hypergraph.JoshuaVisualizationString(self.hg[0]).c_str(), 'utf8')

    def kbest(self, size):
        """hg.kbest(size) -> List of k-best hypotheses in the hypergraph."""
        cdef kbest.KBestDerivations[vector[WordID], kbest.ESentenceTraversal, kbest.NoFilter[vector[int]]]* derivations = new kbest.KBestDerivations[vector[WordID], kbest.ESentenceTraversal, kbest.NoFilter[vector[int]]](self.hg[0], size)
        cdef kbest.KBestDerivations[vector[WordID], kbest.ESentenceTraversal, kbest.NoFilter[vector[int]]].Derivation* derivation
        cdef unsigned k
        try:
            for k in range(size):
                derivation = derivations.LazyKthBest(self.hg.nodes_.size() - 1, k)
                if not derivation: break
                yield unicode(GetString(derivation._yield).c_str(), 'utf8')
        finally:
            del derivations

    def kbest_trees(self, size):
        """hg.kbest_trees(size) -> List of k-best trees in the hypergrapt.NoFilter."""
        cdef kbest.KBestDerivations[vector[WordID], kbest.FTreeTraversal, kbest.NoFilter[vector[int]]]* f_derivations = new kbest.KBestDerivations[vector[WordID], kbest.FTreeTraversal, kbest.NoFilter[vector[int]]](self.hg[0], size)
        cdef kbest.KBestDerivations[vector[WordID], kbest.FTreeTraversal, kbest.NoFilter[vector[int]]].Derivation* f_derivation
        cdef kbest.KBestDerivations[vector[WordID], kbest.ETreeTraversal, kbest.NoFilter[vector[int]]]* e_derivations = new kbest.KBestDerivations[vector[WordID], kbest.ETreeTraversal, kbest.NoFilter[vector[int]]](self.hg[0], size)
        cdef kbest.KBestDerivations[vector[WordID], kbest.ETreeTraversal, kbest.NoFilter[vector[int]]].Derivation* e_derivation
        cdef unsigned k
        try:
            for k in range(size):
                f_derivation = f_derivations.LazyKthBest(self.hg.nodes_.size() - 1, k)
                e_derivation = e_derivations.LazyKthBest(self.hg.nodes_.size() - 1, k)
                if not f_derivation or not e_derivation: break
                f_tree = unicode(GetString(f_derivation._yield).c_str(), 'utf8')
                e_tree = unicode(GetString(e_derivation._yield).c_str(), 'utf8')
                yield (f_tree, e_tree)
        finally:
            del f_derivations
            del e_derivations

    def kbest_features(self, size):
        """hg.kbest_trees(size) -> List of k-best feature vectors in the hypergraph."""
        cdef kbest.KBestDerivations[FastSparseVector[weight_t], kbest.FeatureVectorTraversal, kbest.NoFilter[FastSparseVector[double]]]* derivations = new kbest.KBestDerivations[FastSparseVector[weight_t], kbest.FeatureVectorTraversal, kbest.NoFilter[FastSparseVector[double]]](self.hg[0], size)
        cdef kbest.KBestDerivations[FastSparseVector[weight_t], kbest.FeatureVectorTraversal, kbest.NoFilter[FastSparseVector[double]]].Derivation* derivation
        cdef SparseVector fmap
        cdef unsigned k
        try:
            for k in range(size):
                derivation = derivations.LazyKthBest(self.hg.nodes_.size() - 1, k)
                if not derivation: break
                fmap = SparseVector.__new__(SparseVector)
                fmap.vector = new FastSparseVector[weight_t](derivation._yield)
                yield fmap
        finally:
            del derivations

    def unique_kbest(self, size):
        """hg.kbest(size) -> List of unique k-best hypotheses in the hypergraph."""
        cdef kbest.KBestDerivations[vector[WordID], kbest.ESentenceTraversal, kbest.FilterUnique]* derivations = new kbest.KBestDerivations[vector[WordID], kbest.ESentenceTraversal, kbest.FilterUnique](self.hg[0], size)
        cdef kbest.KBestDerivations[vector[WordID], kbest.ESentenceTraversal, kbest.FilterUnique].Derivation* derivation
        cdef unsigned k
        try:
            for k in range(size):
                derivation = derivations.LazyKthBest(self.hg.nodes_.size() - 1, k)
                if not derivation: break
                yield unicode(GetString(derivation._yield).c_str(), 'utf8')
        finally:
            del derivations

    def unique_kbest_trees(self, size):
        """hg.kbest_trees(size) -> List of unique k-best trees in the hypergraph."""
        cdef kbest.KBestDerivations[vector[WordID], kbest.FTreeTraversal, kbest.FilterUnique]* f_derivations = new kbest.KBestDerivations[vector[WordID], kbest.FTreeTraversal, kbest.FilterUnique](self.hg[0], size)
        cdef kbest.KBestDerivations[vector[WordID], kbest.FTreeTraversal, kbest.FilterUnique].Derivation* f_derivation
        cdef kbest.KBestDerivations[vector[WordID], kbest.ETreeTraversal, kbest.FilterUnique]* e_derivations = new kbest.KBestDerivations[vector[WordID], kbest.ETreeTraversal, kbest.FilterUnique](self.hg[0], size)
        cdef kbest.KBestDerivations[vector[WordID], kbest.ETreeTraversal, kbest.FilterUnique].Derivation* e_derivation
        cdef unsigned k
        try:
            for k in range(size):
                f_derivation = f_derivations.LazyKthBest(self.hg.nodes_.size() - 1, k)
                e_derivation = e_derivations.LazyKthBest(self.hg.nodes_.size() - 1, k)
                if not f_derivation or not e_derivation: break
                f_tree = unicode(GetString(f_derivation._yield).c_str(), 'utf8')
                e_tree = unicode(GetString(e_derivation._yield).c_str(), 'utf8')
                yield (f_tree, e_tree)
        finally:
            del f_derivations
            del e_derivations

    def unique_kbest_features(self, size):
        """hg.kbest_trees(size) -> List of unique k-best feature vectors in the hypergraph."""
        cdef kbest.KBestDerivations[FastSparseVector[weight_t], kbest.FeatureVectorTraversal, kbest.NoFilter[FastSparseVector[double]]]* derivations = new kbest.KBestDerivations[FastSparseVector[weight_t], kbest.FeatureVectorTraversal, kbest.NoFilter[FastSparseVector[double]]](self.hg[0], size)
        cdef kbest.KBestDerivations[FastSparseVector[weight_t], kbest.FeatureVectorTraversal, kbest.NoFilter[FastSparseVector[double]]].Derivation* derivation
        cdef SparseVector fmap
        cdef unsigned k
        try:
            for k in range(size):
                derivation = derivations.LazyKthBest(self.hg.nodes_.size() - 1, k)
                if not derivation: break
                fmap = SparseVector.__new__(SparseVector)
                fmap.vector = new FastSparseVector[weight_t](derivation._yield)
                yield fmap
        finally:
            del derivations

    def sample(self, unsigned n):
        """hg.sample(n) -> Sample of n hypotheses from the hypergraph."""
        cdef vector[hypergraph.Hypothesis]* hypos = new vector[hypergraph.Hypothesis]()
        hypergraph.sample_hypotheses(self.hg[0], n, self._rng(), hypos)
        cdef unsigned k
        try:
            for k in range(hypos.size()):
                yield unicode(GetString(hypos[0][k].words).c_str(), 'utf8')
        finally:
            del hypos
    
    def sample_hypotheses(self, unsigned n):
        """hg.sample_string(n) -> Sample of n hypotheses from the hypergraph.
        Generates (sample_string, dot, fmap)"""
        cdef vector[hypergraph.Hypothesis]* hypos = new vector[hypergraph.Hypothesis]()
        cdef SparseVector fmap = None
        hypergraph.sample_hypotheses(self.hg[0], n, self._rng(), hypos)
        cdef unsigned k
        try:
            for k in range(hypos.size()):
                fmap = SparseVector.__new__(SparseVector)
                fmap.vector = new FastSparseVector[weight_t](hypos[0][k].fmap)
                yield unicode(GetString(hypos[0][k].words).c_str(), 'utf8'), hypos[0][k].model_score.as_float(), fmap
        finally:
            del hypos

    def sample_trees(self, unsigned n):
       """hg.sample_trees(n) -> Sample of n trees from the hypergraph."""
       cdef vector[string]* trees = new vector[string]()
       hypergraph.sample_trees(self.hg[0], n, self._rng(), trees)
       cdef unsigned k
       try:
           for k in range(trees.size()):
               yield unicode(trees[0][k].c_str(), 'utf8')
       finally:
           del trees

    def intersect(self, inp):
        """hg.intersect(Lattice/string): Intersect the hypergraph with the provided reference."""
        cdef Lattice lat
        if isinstance(inp, Lattice):
            lat = <Lattice> inp
        elif isinstance(inp, basestring):
            lat = Lattice(inp)
        else:
            raise TypeError('cannot intersect hypergraph with %s' % type(inp))
        return hypergraph.Intersect(lat.lattice[0], self.hg)

    def prune(self, beam_alpha=0, density=0, **kwargs):
        """hg.prune(beam_alpha=0, density=0): Prune the hypergraph.
        beam_alpha: use beam pruning
        density: use density pruning"""
        cdef hypergraph.EdgeMask* preserve_mask = NULL
        if 'csplit_preserve_full_word' in kwargs:
             preserve_mask = new hypergraph.EdgeMask(self.hg.edges_.size())
             preserve_mask[0][hypergraph.GetFullWordEdgeIndex(self.hg[0])] = True
        self.hg.PruneInsideOutside(beam_alpha, density, preserve_mask, False, 1, False)
        if preserve_mask:
            del preserve_mask

    def lattice(self): # TODO direct hg -> lattice conversion in cdec
        """hg.lattice() -> Lattice corresponding to the hypergraph."""
        cdef bytes plf = hypergraph.AsPLF(self.hg[0], True).c_str()
        return Lattice(eval(plf))

    def plf(self):
        """hg.plf() -> Lattice PLF representation corresponding to the hypergraph."""
        return bytes(hypergraph.AsPLF(self.hg[0], True).c_str())

    def reweight(self, weights):
        """hg.reweight(SparseVector/DenseVector): Reweight the hypergraph with a new vector."""
        if isinstance(weights, SparseVector):
            self.hg.Reweight((<SparseVector> weights).vector[0])
        elif isinstance(weights, DenseVector):
            self.hg.Reweight((<DenseVector> weights).vector[0])
        else:
            raise TypeError('cannot reweight hypergraph with %s' % type(weights))

    property edges:
        def __get__(self):
            cdef unsigned i
            for i in range(self.hg.edges_.size()):
                yield HypergraphEdge().init(self.hg, i)

    property nodes:
        def __get__(self):
            cdef unsigned i
            for i in range(self.hg.nodes_.size()):
                yield HypergraphNode().init(self.hg, i)

    property goal:
        def __get__(self):
            return HypergraphNode().init(self.hg, self.hg.GoalNode())

    property npaths:
        def __get__(self):
            return self.hg.NumberOfPaths()

    def inside_outside(self):
        """hg.inside_outside() -> SparseVector with inside-outside scores for each feature."""
        cdef FastSparseVector[prob_t]* result = new FastSparseVector[prob_t]()
        cdef prob_t z = hypergraph.InsideOutside(self.hg[0], result)
        result[0] /= z
        cdef SparseVector vector = SparseVector.__new__(SparseVector)
        vector.vector = new FastSparseVector[double]()
        cdef FastSparseVector[prob_t].const_iterator* it = new FastSparseVector[prob_t].const_iterator(result[0], False)
        cdef unsigned i
        for i in range(result.size()):
            vector.vector.set_value(it[0].ptr().first, log(it[0].ptr().second))
            pinc(it[0]) # ++it
        del it
        del result
        return vector

cdef class HypergraphEdge:
    cdef hypergraph.Hypergraph* hg
    cdef hypergraph.HypergraphEdge* edge
    cdef public TRule trule

    cdef init(self, hypergraph.Hypergraph* hg, unsigned i):
        self.hg = hg
        self.edge = &hg.edges_[i]
        self.trule = TRule.__new__(TRule)
        self.trule.rule = new shared_ptr[grammar.TRule](self.edge.rule_)
        return self

    def __len__(self):
        return self.edge.tail_nodes_.size()

    property head_node:
        def __get__(self):
            return HypergraphNode().init(self.hg, self.edge.head_node_)

    property tail_nodes:
        def __get__(self):
            cdef unsigned i
            for i in range(self.edge.tail_nodes_.size()):
                yield HypergraphNode().init(self.hg, self.edge.tail_nodes_[i])

    property span:
        def __get__(self):
            return (self.edge.i_, self.edge.j_)

    property src_span:
        def __get__(self):
            return (self.edge.prev_i_, self.edge.prev_j_)

    property feature_values:
        def __get__(self):
            cdef SparseVector vector = SparseVector.__new__(SparseVector)
            vector.vector = new FastSparseVector[double](self.edge.feature_values_)
            return vector

    property prob:
        def __get__(self):
            return self.edge.edge_prob_.as_float()

    def __richcmp__(HypergraphEdge x, HypergraphEdge y, int op):
        if op == 2: # ==
            return x.edge == y.edge
        elif op == 3: # !=
            return not (x == y)
        raise NotImplemented('comparison not implemented for HypergraphEdge')

cdef class HypergraphNode:
    cdef hypergraph.Hypergraph* hg
    cdef hypergraph.HypergraphNode* node

    cdef init(self, hypergraph.Hypergraph* hg, unsigned i):
        self.hg = hg
        self.node = &hg.nodes_[i]
        return self

    property id:
        def __get__(self):
            return self.node.id_

    property in_edges:
        def __get__(self):
            cdef unsigned i
            for i in range(self.node.in_edges_.size()):
                yield HypergraphEdge().init(self.hg, self.node.in_edges_[i])

    property out_edges:
        def __get__(self):
            cdef unsigned i
            for i in range(self.node.out_edges_.size()):
                yield HypergraphEdge().init(self.hg, self.node.out_edges_[i])

    property span:
        def __get__(self):
            return next(self.in_edges).span

    property cat:
        def __get__(self):
            if self.node.cat_:
                return str(TDConvert(-self.node.cat_).c_str())

    def __richcmp__(HypergraphNode x, HypergraphNode y, int op):
        if op == 2: # ==
            return x.node == y.node
        elif op == 3: # !=
            return not (x == y)
        raise NotImplemented('comparison not implemented for HypergraphNode')