Add beam decoding to parser, to allow NER uncertainties

spacy/syntax/arc_eager.pyx

@@ -10,6 +10,8 @@ from libc.stdint cimport uint32_t
 from libc.string cimport memcpy
 from cymem.cymem cimport Pool
 from collections import OrderedDict
+from thinc.extra.search cimport Beam
+import numpy
 
 from .stateclass cimport StateClass
 from ._state cimport StateC, is_space_token
@@ -510,3 +512,23 @@ cdef class ArcEager(TransitionSystem):
                     "State at failure:\n"
                     "%s" % (self.n_moves, stcls.print_state(gold.words)))
         assert n_gold >= 1
+
+    def get_beam_annot(self, Beam beam):
+        length = (<StateClass>beam.at(0)).c.length
+        heads = [{} for _ in range(length)]
+        deps = [{} for _ in range(length)]
+        probs = beam.probs
+        for i in range(beam.size):
+            stcls = <StateClass>beam.at(i)
+            self.finalize_state(stcls.c)
+            if stcls.is_final():
+                prob = probs[i]
+                for j in range(stcls.c.length):
+                    head = j + stcls.c._sent[j].head
+                    dep = stcls.c._sent[j].dep
+                    heads[j].setdefault(head, 0.0)
+                    heads[j][head] += prob
+                    deps[j].setdefault(dep, 0.0)
+                    deps[j][dep] += prob
+        return heads, deps
+

spacy/syntax/ner.pyx

@@ -2,7 +2,10 @@
 from __future__ import unicode_literals
 
 from thinc.typedefs cimport weight_t
+from thinc.extra.search cimport Beam
 from collections import OrderedDict
+import numpy
+from thinc.neural.ops import NumpyOps
 
 from .stateclass cimport StateClass
 from ._state cimport StateC
@@ -122,6 +125,22 @@ cdef class BiluoPushDown(TransitionSystem):
             gold.c.ner[i] = self.lookup_transition(gold.ner[i])
         return gold
 
+    def get_beam_annot(self, Beam beam):
+        entities = {}
+        probs = beam.probs
+        for i in range(beam.size):
+            stcls = <StateClass>beam.at(i)
+            if stcls.is_final():
+                self.finalize_state(stcls.c)
+                prob = probs[i]
+                for j in range(stcls.c._e_i):
+                    start = stcls.c._ents[j].start
+                    end = stcls.c._ents[j].end
+                    label = stcls.c._ents[j].label
+                    entities.setdefault((start, end, label), 0.0)
+                    entities[(start, end, label)] += prob
+        return entities
+
     cdef Transition lookup_transition(self, object name) except *:
         cdef attr_t label
         if name == '-' or name == None:

spacy/syntax/nn_parser.pyx

@@ -29,6 +29,7 @@ from thinc.linear.avgtron cimport AveragedPerceptron
 from thinc.linalg cimport VecVec
 from thinc.structs cimport SparseArrayC, FeatureC, ExampleC
 from thinc.extra.eg cimport Example
+from thinc.extra.search cimport Beam
 
 from cymem.cymem cimport Pool, Address
 from murmurhash.mrmr cimport hash64
@@ -110,7 +111,6 @@ cdef class precompute_hiddens:
         self.nO = cached.shape[2]
         self.nP = getattr(lower_model, 'nP', 1)
         self.ops = lower_model.ops
-        self._features = numpy.zeros((batch_size, self.nO*self.nP), dtype='f')
         self._is_synchronized = False
         self._cuda_stream = cuda_stream
         self._cached = cached
@@ -127,13 +127,12 @@ cdef class precompute_hiddens:
         return self.begin_update(X)[0]
 
     def begin_update(self, token_ids, drop=0.):
-        self._features.fill(0)
+        cdef np.ndarray state_vector = numpy.zeros((token_ids.shape[0], self.nO*self.nP), dtype='f')
         # This is tricky, but (assuming GPU available);
         # - Input to forward on CPU
         # - Output from forward on CPU
         # - Input to backward on GPU!
         # - Output from backward on GPU
-        cdef np.ndarray state_vector = self._features[:len(token_ids)]
         bp_hiddens = self._bp_hiddens
 
         feat_weights = self.get_feat_weights()
@@ -305,7 +304,7 @@ cdef class Parser:
     def __reduce__(self):
         return (Parser, (self.vocab, self.moves, self.model), None, None)
 
-    def __call__(self, Doc doc):
+    def __call__(self, Doc doc, beam_width=None, beam_density=None):
         """
         Apply the parser or entity recognizer, setting the annotations onto the Doc object.
 
@@ -314,11 +313,26 @@ cdef class Parser:
         Returns:
             None
         """
+        if beam_width is None:
+            beam_width = self.cfg.get('beam_width', 1)
+        if beam_density is None:
+            beam_density = self.cfg.get('beam_density', 0.001)
+        cdef Beam beam
+        if beam_width == 1:
             states = self.parse_batch([doc], [doc.tensor])
             self.set_annotations([doc], states)
             return doc
+        else:
+            beam = self.beam_parse([doc], [doc.tensor],
+                        beam_width=beam_width, beam_density=beam_density)[0]
+            output = self.moves.get_beam_annot(beam)
+            state = <StateClass>beam.at(0)
+            self.set_annotations([doc], [state])
+            _cleanup(beam)
+            return output
 
-    def pipe(self, docs, int batch_size=1000, int n_threads=2):
+    def pipe(self, docs, int batch_size=1000, int n_threads=2,
+             beam_width=1, beam_density=0.001):
         """
         Process a stream of documents.
 
@@ -336,7 +350,11 @@ cdef class Parser:
         for docs in cytoolz.partition_all(batch_size, docs):
             docs = list(docs)
             tokvecs = [d.tensor for d in docs]
+            if beam_width == 1:
                 parse_states = self.parse_batch(docs, tokvecs)
+            else:
+                parse_states = self.beam_parse(docs, tokvecs,
+                                    beam_width=beam_width, beam_density=beam_density)
             self.set_annotations(docs, parse_states)
             yield from docs
 
@@ -404,6 +422,45 @@ cdef class Parser:
                     next_step.push_back(st)
         return states
 
+    def beam_parse(self, docs, tokvecses, int beam_width=8, float beam_density=0.001):
+        cdef Beam beam
+        cdef np.ndarray scores
+        cdef Doc doc
+        cdef int nr_class = self.moves.n_moves
+        cdef StateClass stcls, output
+        tokvecs = self.model[0].ops.flatten(tokvecses)
+        cuda_stream = get_cuda_stream()
+        state2vec, vec2scores = self.get_batch_model(len(docs), tokvecs,
+                                                     cuda_stream, 0.0)
+        beams = []
+        cdef int offset = 0
+        for doc in docs:
+            beam = Beam(nr_class, beam_width, min_density=beam_density)
+            beam.initialize(self.moves.init_beam_state, doc.length, doc.c)
+            for i in range(beam.width):
+                stcls = <StateClass>beam.at(i)
+                stcls.c.offset = offset
+            offset += len(doc)
+            beam.check_done(_check_final_state, NULL)
+            while not beam.is_done:
+                states = []
+                for i in range(beam.size):
+                    stcls = <StateClass>beam.at(i)
+                    states.append(stcls)
+                token_ids = self.get_token_ids(states)
+                vectors = state2vec(token_ids)
+                scores = vec2scores(vectors)
+                for i in range(beam.size):
+                    stcls = <StateClass>beam.at(i)
+                    if not stcls.is_final():
+                        self.moves.set_valid(beam.is_valid[i], stcls.c)
+                        for j in range(nr_class):
+                            beam.scores[i][j] = scores[i, j]
+                beam.advance(_transition_state, _hash_state, <void*>self.moves.c)
+                beam.check_done(_check_final_state, NULL)
+            beams.append(beam)
+        return beams
+
     cdef void _parse_step(self, StateC* state,
             const float* feat_weights,
             int nr_class, int nr_feat, int nr_piece) nogil:
@@ -560,6 +617,7 @@ cdef class Parser:
                                           dtype='i', order='C')
         c_ids = <int*>ids.data
         for i, state in enumerate(states):
+            if not state.is_final():
                 state.c.set_context_tokens(c_ids, n_tokens)
             c_ids += ids.shape[1]
         return ids
@@ -762,3 +820,30 @@ cdef int _arg_max_clas(const weight_t* scores, int move, const Transition* actio
             mode = i
             score = scores[i]
     return mode
+
+
+# These are passed as callbacks to thinc.search.Beam
+cdef int _transition_state(void* _dest, void* _src, class_t clas, void* _moves) except -1:
+    dest = <StateClass>_dest
+    src = <StateClass>_src
+    moves = <const Transition*>_moves
+    dest.clone(src)
+    moves[clas].do(dest.c, moves[clas].label)
+
+
+cdef int _check_final_state(void* _state, void* extra_args) except -1:
+    return (<StateClass>_state).is_final()
+
+
+def _cleanup(Beam beam):
+    for i in range(beam.width):
+        Py_XDECREF(<PyObject*>beam._states[i].content)
+        Py_XDECREF(<PyObject*>beam._parents[i].content)
+
+
+cdef hash_t _hash_state(void* _state, void* _) except 0:
+    state = <StateClass>_state
+    if state.c.is_final():
+        return 1
+    else:
+        return state.c.hash()

spacy/syntax/transition_system.pyx

@@ -137,6 +137,10 @@ cdef class TransitionSystem:
                 "the entity recognizer\n"
                 "The transition system has %d actions." % (self.n_moves))
 
+    def get_class_name(self, int clas):
+        act = self.c[clas]
+        return self.move_name(act.move, act.label)
+
     def add_action(self, int action, label_name):
         cdef attr_t label_id
         if not isinstance(label_name, int):