Introduces a bulk merge function, in order to solve issue #653 (#2696)

* Fix comment

* Introduce bulk merge to increase performance on many span merges

* Sign contributor agreement

* Implement pull request suggestions

.github/contributors/grivaz.md

@@ -0,0 +1,106 @@
+# spaCy contributor agreement
+
+This spaCy Contributor Agreement (**"SCA"**) is based on the
+[Oracle Contributor Agreement](http://www.oracle.com/technetwork/oca-405177.pdf).
+The SCA applies to any contribution that you make to any product or project
+managed by us (the **"project"**), and sets out the intellectual property rights
+you grant to us in the contributed materials. The term **"us"** shall mean
+[ExplosionAI UG (haftungsbeschränkt)](https://explosion.ai/legal). The term
+**"you"** shall mean the person or entity identified below.
+
+If you agree to be bound by these terms, fill in the information requested
+below and include the filled-in version with your first pull request, under the
+folder [`.github/contributors/`](/.github/contributors/). The name of the file
+should be your GitHub username, with the extension `.md`. For example, the user
+example_user would create the file `.github/contributors/example_user.md`.
+
+Read this agreement carefully before signing. These terms and conditions
+constitute a binding legal agreement.
+
+## Contributor Agreement
+
+1. The term "contribution" or "contributed materials" means any source code,
+object code, patch, tool, sample, graphic, specification, manual,
+documentation, or any other material posted or submitted by you to the project.
+
+2. With respect to any worldwide copyrights, or copyright applications and
+registrations, in your contribution:
+
+    * you hereby assign to us joint ownership, and to the extent that such
+    assignment is or becomes invalid, ineffective or unenforceable, you hereby
+    grant to us a perpetual, irrevocable, non-exclusive, worldwide, no-charge,
+    royalty-free, unrestricted license to exercise all rights under those
+    copyrights. This includes, at our option, the right to sublicense these same
+    rights to third parties through multiple levels of sublicensees or other
+    licensing arrangements;
+
+    * you agree that each of us can do all things in relation to your
+    contribution as if each of us were the sole owners, and if one of us makes
+    a derivative work of your contribution, the one who makes the derivative
+    work (or has it made will be the sole owner of that derivative work;
+
+    * you agree that you will not assert any moral rights in your contribution
+    against us, our licensees or transferees;
+
+    * you agree that we may register a copyright in your contribution and
+    exercise all ownership rights associated with it; and
+
+    * you agree that neither of us has any duty to consult with, obtain the
+    consent of, pay or render an accounting to the other for any use or
+    distribution of your contribution.
+
+3. With respect to any patents you own, or that you can license without payment
+to any third party, you hereby grant to us a perpetual, irrevocable,
+non-exclusive, worldwide, no-charge, royalty-free license to:
+
+    * make, have made, use, sell, offer to sell, import, and otherwise transfer
+    your contribution in whole or in part, alone or in combination with or
+    included in any product, work or materials arising out of the project to
+    which your contribution was submitted, and
+
+    * at our option, to sublicense these same rights to third parties through
+    multiple levels of sublicensees or other licensing arrangements.
+
+4. Except as set out above, you keep all right, title, and interest in your
+contribution. The rights that you grant to us under these terms are effective
+on the date you first submitted a contribution to us, even if your submission
+took place before the date you sign these terms.
+
+5. You covenant, represent, warrant and agree that:
+
+    * Each contribution that you submit is and shall be an original work of
+    authorship and you can legally grant the rights set out in this SCA;
+
+    * to the best of your knowledge, each contribution will not violate any
+    third party's copyrights, trademarks, patents, or other intellectual
+    property rights; and
+
+    * each contribution shall be in compliance with U.S. export control laws and
+    other applicable export and import laws. You agree to notify us if you
+    become aware of any circumstance which would make any of the foregoing
+    representations inaccurate in any respect. We may publicly disclose your
+    participation in the project, including the fact that you have signed the SCA.
+
+6. This SCA is governed by the laws of the State of California and applicable
+U.S. Federal law. Any choice of law rules will not apply.
+
+7. Please place an “x” on one of the applicable statement below. Please do NOT
+mark both statements:
+
+    * [x] I am signing on behalf of myself as an individual and no other person
+    or entity, including my employer, has or will have rights with respect to my
+    contributions.
+
+    * [ ] I am signing on behalf of my employer or a legal entity and I have the
+    actual authority to contractually bind that entity.
+
+## Contributor Details
+
+| Field                          | Entry                |
+|------------------------------- | -------------------- |
+| Name                           |C. Grivaz                   |
+| Company name (if applicable)   |                      |
+| Title or role (if applicable)  |                      |
+| Date                           |08.22.2018                  |
+| GitHub username                |grivaz               |
+| Website (optional)             |                      |

spacy/errors.py

@@ -249,6 +249,7 @@ class Errors(object):
             "error. Are you writing to a default function argument?")
     E096 = ("Invalid object passed to displaCy: Can only visualize Doc or "
              "Span objects, or dicts if set to manual=True.")
+    E097 = ("Can't merge non-disjoint spans. '{token}' is already part of tokens to merge")
 
 
 @add_codes

spacy/tests/doc/test_doc_api.py

@@ -5,6 +5,7 @@ from ..util import get_doc
 from ...tokens import Doc
 from ...vocab import Vocab
 from ...attrs import LEMMA
+from ...tokens import Span
 
 import pytest
 import numpy
@@ -156,6 +157,23 @@ def test_doc_api_merge(en_tokenizer):
     assert doc[7].text == 'all night'
     assert doc[7].text_with_ws == 'all night'
 
+    # merge both with bulk merge
+    doc = en_tokenizer(text)
+    assert len(doc) == 9
+    with doc.retokenize() as retokenizer:
+        retokenizer.merge(doc[4: 7], attrs={'tag':'NAMED', 'lemma':'LEMMA',
+              'ent_type':'TYPE'})
+        retokenizer.merge(doc[7: 9], attrs={'tag':'NAMED', 'lemma':'LEMMA',
+              'ent_type':'TYPE'})
+
+    assert len(doc) == 6
+    assert doc[4].text == 'the beach boys'
+    assert doc[4].text_with_ws == 'the beach boys '
+    assert doc[4].tag_ == 'NAMED'
+    assert doc[5].text == 'all night'
+    assert doc[5].text_with_ws == 'all night'
+    assert doc[5].tag_ == 'NAMED'
+
 
 def test_doc_api_merge_children(en_tokenizer):
     """Test that attachments work correctly after merging."""

spacy/tests/doc/test_span_merge.py

@@ -4,6 +4,7 @@ from __future__ import unicode_literals
 from ..util import get_doc
 from ...vocab import Vocab
 from ...tokens import Doc
+from ...tokens import Span
 
 import pytest
 
@@ -16,16 +17,8 @@ def test_spans_merge_tokens(en_tokenizer):
     assert len(doc) == 4
     assert doc[0].head.text == 'Angeles'
     assert doc[1].head.text == 'start'
-    doc.merge(0, len('Los Angeles'), tag='NNP', lemma='Los Angeles', ent_type='GPE')
+    with doc.retokenize() as retokenizer:
-    assert len(doc) == 3
+        retokenizer.merge(doc[0 : 2], attrs={'tag':'NNP', 'lemma':'Los Angeles', 'ent_type':'GPE'})
-    assert doc[0].text == 'Los Angeles'
-    assert doc[0].head.text == 'start'
-
-    doc = get_doc(tokens.vocab, [t.text for t in tokens], heads=heads)
-    assert len(doc) == 4
-    assert doc[0].head.text == 'Angeles'
-    assert doc[1].head.text == 'start'
-    doc.merge(0, len('Los Angeles'), tag='NNP', lemma='Los Angeles', label='GPE')
     assert len(doc) == 3
     assert doc[0].text == 'Los Angeles'
     assert doc[0].head.text == 'start'
@@ -38,8 +31,8 @@ def test_spans_merge_heads(en_tokenizer):
     doc = get_doc(tokens.vocab, [t.text for t in tokens], heads=heads)
 
     assert len(doc) == 8
-    doc.merge(doc[3].idx, doc[4].idx + len(doc[4]), tag=doc[4].tag_,
+    with doc.retokenize() as retokenizer:
-              lemma='pilates class', ent_type='O')
+        retokenizer.merge(doc[3 : 5], attrs={'tag':doc[4].tag_, 'lemma':'pilates class', 'ent_type':'O'})
     assert len(doc) == 7
     assert doc[0].head.i == 1
     assert doc[1].head.i == 1
@@ -48,6 +41,14 @@ def test_spans_merge_heads(en_tokenizer):
     assert doc[4].head.i in [1, 3]
     assert doc[5].head.i == 4
 
+def test_spans_merge_non_disjoint(en_tokenizer):
+    text = "Los Angeles start."
+    tokens = en_tokenizer(text)
+    doc = get_doc(tokens.vocab, [t.text for t in tokens])
+    with pytest.raises(ValueError):
+        with doc.retokenize() as retokenizer:
+            retokenizer.merge(doc[0: 2], attrs={'tag': 'NNP', 'lemma': 'Los Angeles', 'ent_type': 'GPE'})
+            retokenizer.merge(doc[0: 1], attrs={'tag': 'NNP', 'lemma': 'Los Angeles', 'ent_type': 'GPE'})
 
 def test_span_np_merges(en_tokenizer):
     text = "displaCy is a parse tool built with Javascript"
@@ -111,6 +112,25 @@ def test_spans_entity_merge_iob():
     assert doc[0].ent_iob_ == "B"
     assert doc[1].ent_iob_ == "I"
 
+    words = ["a", "b", "c", "d", "e", "f", "g", "h", "i"]
+    doc = Doc(Vocab(), words=words)
+    doc.ents = [(doc.vocab.strings.add('ent-de'), 3, 5),
+                (doc.vocab.strings.add('ent-fg'), 5, 7)]
+    assert doc[3].ent_iob_ == "B"
+    assert doc[4].ent_iob_ == "I"
+    assert doc[5].ent_iob_ == "B"
+    assert doc[6].ent_iob_ == "I"
+    with doc.retokenize() as retokenizer:
+        retokenizer.merge(doc[2 : 4])
+        retokenizer.merge(doc[4 : 6])
+        retokenizer.merge(doc[7 : 9])
+    for token in doc:
+        print(token)
+        print(token.ent_iob)
+    assert len(doc) == 6
+    assert doc[3].ent_iob_ == "B"
+    assert doc[4].ent_iob_ == "I"
+
 
 def test_spans_sentence_update_after_merge(en_tokenizer):
     text = "Stewart Lee is a stand up comedian. He lives in England and loves Joe Pasquale."

spacy/tokens/_retokenize.pyx

@@ -5,6 +5,9 @@
 from __future__ import unicode_literals
 
 from libc.string cimport memcpy, memset
+from libc.stdlib cimport malloc, free
+
+from cymem.cymem cimport Pool
 
 from .doc cimport Doc, set_children_from_heads, token_by_start, token_by_end
 from .span cimport Span
@@ -14,24 +17,31 @@ from ..structs cimport LexemeC, TokenC
 from ..attrs cimport TAG
 from ..attrs import intify_attrs
 from ..util import SimpleFrozenDict
-
+from ..errors import Errors
 
 cdef class Retokenizer:
     """Helper class for doc.retokenize() context manager."""
     cdef Doc doc
     cdef list merges
     cdef list splits
+    cdef set tokens_to_merge
     def __init__(self, doc):
         self.doc = doc
         self.merges = []
         self.splits = []
+        self.tokens_to_merge = set()
 
     def merge(self, Span span, attrs=SimpleFrozenDict()):
         """Mark a span for merging. The attrs will be applied to the resulting
         token.
         """
+        for token in span:
+            if token.i in self.tokens_to_merge:
+                raise ValueError(Errors.E097.format(token=repr(token)))
+            self.tokens_to_merge.add(token.i)
+
         attrs = intify_attrs(attrs, strings_map=self.doc.vocab.strings)
-        self.merges.append((span.start_char, span.end_char, attrs))
+        self.merges.append((span, attrs))
 
     def split(self, Token token, orths, attrs=SimpleFrozenDict()):
         """Mark a Token for splitting, into the specified orths. The attrs
@@ -47,20 +57,22 @@ cdef class Retokenizer:
 
     def __exit__(self, *args):
         # Do the actual merging here
-        for start_char, end_char, attrs in self.merges:
+        if len(self.merges) > 1:
-            start = token_by_start(self.doc.c, self.doc.length, start_char)
+            _bulk_merge(self.doc, self.merges)
-            end = token_by_end(self.doc.c, self.doc.length, end_char)
+        elif len(self.merges) == 1:
-            _merge(self.doc, start, end+1, attrs)
+            (span, attrs) = self.merges[0]
+            start = span.start
+            end = span.end
+            _merge(self.doc, start, end, attrs)
+
         for start_char, orths, attrs in self.splits:
             raise NotImplementedError
 
-
 def _merge(Doc doc, int start, int end, attributes):
     """Retokenize the document, such that the span at
     `doc.text[start_idx : end_idx]` is merged into a single token. If
     `start_idx` and `end_idx `do not mark start and end token boundaries,
     the document remains unchanged.
-
     start_idx (int): Character index of the start of the slice to merge.
     end_idx (int): Character index after the end of the slice to merge.
     **attributes: Attributes to assign to the merged token. By default,
@@ -131,3 +143,139 @@ def _merge(Doc doc, int start, int end, attributes):
     # Clear the cached Python objects
     # Return the merged Python object
     return doc[start]
+
+def _bulk_merge(Doc doc, merges):
+    """Retokenize the document, such that the spans described in 'merges'
+     are merged into a single token. This method assumes that the merges
+     are in the same order at which they appear in the doc, and that merges
+     do not intersect each other in any way.
+
+    merges: Tokens to merge, and corresponding attributes to assign to the
+        merged token. By default, attributes are inherited from the
+        syntactic root of the span.
+    RETURNS (Token): The first newly merged token.
+    """
+    cdef Span span
+    cdef const LexemeC* lex
+    cdef Pool mem = Pool()
+    tokens = <TokenC**>mem.alloc(len(merges), sizeof(TokenC))
+    spans = []
+
+    def _get_start(merge):
+        return merge[0].start
+    merges.sort(key=_get_start)
+
+    for merge_index, (span, attributes) in enumerate(merges):
+        start = span.start
+        end = span.end
+        spans.append(span)
+
+        # House the new merged token where it starts
+        token = &doc.c[start]
+
+        tokens[merge_index] = token
+
+        # Assign attributes
+        for attr_name, attr_value in attributes.items():
+            if attr_name == TAG:
+                doc.vocab.morphology.assign_tag(token, attr_value)
+            else:
+                Token.set_struct_attr(token, attr_name, attr_value)
+
+
+    # Memorize span roots and sets dependencies of the newly merged
+    # tokens to the dependencies of their roots.
+    span_roots = []
+    for i, span in enumerate(spans):
+        span_roots.append(span.root.i)
+        tokens[i].dep = span.root.dep
+
+    # We update token.lex after keeping span root and dep, since
+    # setting token.lex will change span.start and span.end properties
+    # as it modifies the character offsets in the doc
+    for token_index in range(len(merges)):
+        new_orth = ''.join([t.text_with_ws for t in spans[token_index]])
+        if spans[token_index][-1].whitespace_:
+            new_orth = new_orth[:-len(spans[token_index][-1].whitespace_)]
+        lex = doc.vocab.get(doc.mem, new_orth)
+        tokens[token_index].lex = lex
+        # We set trailing space here too
+        tokens[token_index].spacy = doc.c[spans[token_index].end-1].spacy
+
+    # Begin by setting all the head indices to absolute token positions
+    # This is easier to work with for now than the offsets
+    # Before thinking of something simpler, beware the case where a
+    # dependency bridges over the entity. Here the alignment of the
+    # tokens changes.
+    for i in range(doc.length):
+        doc.c[i].head += i
+
+    # Set the head of the merged token from the Span
+    for i in range(len(merges)):
+        tokens[i].head = doc.c[span_roots[i]].head
+
+    # Adjust deps before shrinking tokens
+    # Tokens which point into the merged token should now point to it
+    # Subtract the offset from all tokens which point to >= end
+    offsets = []
+    current_span_index = 0
+    current_offset = 0
+    for i in range(doc.length):
+        if current_span_index < len(spans) and i == spans[current_span_index].end:
+            #last token was the last of the span
+            current_offset += (spans[current_span_index].end - spans[current_span_index].start) -1
+            current_span_index += 1
+
+        if current_span_index < len(spans) and \
+                spans[current_span_index].start <= i < spans[current_span_index].end:
+            offsets.append(spans[current_span_index].start - current_offset)
+        else:
+            offsets.append(i - current_offset)
+
+    for i in range(doc.length):
+        doc.c[i].head = offsets[doc.c[i].head]
+
+    # Now compress the token array
+    offset = 0
+    in_span = False
+    span_index = 0
+    for i in range(doc.length):
+        if in_span and i == spans[span_index].end:
+            # First token after a span
+            in_span = False
+            span_index += 1
+        if span_index < len(spans) and i == spans[span_index].start:
+            # First token in a span
+            doc.c[i - offset] = doc.c[i] # move token to its place
+            offset += (spans[span_index].end - spans[span_index].start) - 1
+            in_span = True
+        if not in_span:
+            doc.c[i - offset] = doc.c[i] # move token to its place
+
+    for i in range(doc.length - offset, doc.length):
+        memset(&doc.c[i], 0, sizeof(TokenC))
+        doc.c[i].lex = &EMPTY_LEXEME
+    doc.length -= offset
+
+    # ...And, set heads back to a relative position
+    for i in range(doc.length):
+        doc.c[i].head -= i
+
+    # Set the left/right children, left/right edges
+    set_children_from_heads(doc.c, doc.length)
+
+    # Make sure ent_iob remains consistent
+    for (span, _) in merges:
+        if(span.end < len(offsets)):
+        #if it's not the last span
+            token_after_span_position = offsets[span.end]
+            if doc.c[token_after_span_position].ent_iob == 1\
+                    and doc.c[token_after_span_position - 1].ent_iob in (0, 2):
+                if doc.c[token_after_span_position - 1].ent_type == doc.c[token_after_span_position].ent_type:
+                    doc.c[token_after_span_position - 1].ent_iob = 3
+                else:
+                    # If they're not the same entity type, let them be two entities
+                    doc.c[token_after_span_position].ent_iob = 3
+
+    # Return the merged Python object
+    return doc[spans[0].start]

spacy/tokens/doc.pyx

@@ -884,6 +884,28 @@ cdef class Doc:
         '''
         return Retokenizer(self)
 
+    def _bulk_merge(self, spans, attributes):
+        """Retokenize the document, such that the spans given as arguments
+         are merged into single tokens. The spans need to be in document
+         order, and no span intersection is allowed.
+
+        spans (Span[]): Spans to merge, in document order, with all span
+            intersections empty. Cannot be emty.
+        attributes (Dictionary[]): Attributes to assign to the merged tokens. By default,
+            must be the same lenghth as spans, emty dictionaries are allowed.
+            attributes are inherited from the syntactic root of the span.
+        RETURNS (Token): The first newly merged token.
+        """
+        cdef unicode tag, lemma, ent_type
+
+        assert len(attributes) == len(spans), "attribute length should be equal to span length" + str(len(attributes)) +\
+                                              str(len(spans))
+        with self.retokenize() as retokenizer:
+            for i, span in enumerate(spans):
+                fix_attributes(self, attributes[i])
+                remove_label_if_necessary(attributes[i])
+                retokenizer.merge(span, attributes[i])
+
     def merge(self, int start_idx, int end_idx, *args, **attributes):
         """Retokenize the document, such that the span at
         `doc.text[start_idx : end_idx]` is merged into a single token. If
@@ -905,20 +927,12 @@ cdef class Doc:
             attributes[LEMMA] = lemma
             attributes[ENT_TYPE] = ent_type
         elif not args:
-            if 'label' in attributes and 'ent_type' not in attributes:
+            fix_attributes(self, attributes)
-                if isinstance(attributes['label'], int):
-                    attributes[ENT_TYPE] = attributes['label']
-                else:
-                    attributes[ENT_TYPE] = self.vocab.strings[attributes['label']]
-            if 'ent_type' in attributes:
-                attributes[ENT_TYPE] = attributes['ent_type']
         elif args:
             raise ValueError(Errors.E034.format(n_args=len(args),
                                                 args=repr(args),
                                                 kwargs=repr(attributes)))
-        # More deprecated attribute handling =/
+        remove_label_if_necessary(attributes)
-        if 'label' in attributes:
-            attributes['ent_type'] = attributes.pop('label')
 
         attributes = intify_attrs(attributes, strings_map=self.vocab.strings)
 
@@ -1034,3 +1048,17 @@ def unpickle_doc(vocab, hooks_and_data, bytes_data):
 
 
 copy_reg.pickle(Doc, pickle_doc, unpickle_doc)
+
+def remove_label_if_necessary(attributes):
+    # More deprecated attribute handling =/
+    if 'label' in attributes:
+        attributes['ent_type'] = attributes.pop('label')
+
+def fix_attributes(doc, attributes):
+    if 'label' in attributes and 'ent_type' not in attributes:
+        if isinstance(attributes['label'], int):
+            attributes[ENT_TYPE] = attributes['label']
+        else:
+            attributes[ENT_TYPE] = doc.vocab.strings[attributes['label']]
+    if 'ent_type' in attributes:
+        attributes[ENT_TYPE] = attributes['ent_type']