Update ner_spancat_compare for newer spacy, doc/vocab issues (#191)

* Rename component to `transfer-ent` to avoid `.` in factory name (for
  spacy v3.4.2+)
* In convert script, make sure all spans belong to the same doc object
  when initializing span groups. Otherwise there can be cases where the
  span label is not in the doc vocab.

experimental/ner_spancat_compare/configs/ner_assemble.cfg

@@ -55,23 +55,23 @@ component = "ner"
 # Then, we instantiate a component from our `transfer-ent` factory so that
 # we can move the entities in doc.ents into doc.spans.
 [components.transfer-dna]
-factory = "transfer-ent.v1"
+factory = "transfer-ent"
 span_key = ${paths.spans_key}
 
 [components.transfer-rna]
-factory = "transfer-ent.v1"
+factory = "transfer-ent"
 span_key = ${paths.spans_key}
 
 [components.transfer-cell-line]
-factory = "transfer-ent.v1"
+factory = "transfer-ent"
 span_key = ${paths.spans_key}
 
 [components.transfer-cell-type]
-factory = "transfer-ent.v1"
+factory = "transfer-ent"
 span_key = ${paths.spans_key}
 
 [components.transfer-protein]
-factory = "transfer-ent.v1"
+factory = "transfer-ent"
 span_key = ${paths.spans_key}
 
 
@@ -157,4 +157,4 @@ after_init = null
 
 [initialize.components]
 
-[initialize.tokenizer]
+[initialize.tokenizer]

experimental/ner_spancat_compare/scripts/convert.py

@@ -2,7 +2,7 @@
 from typing import List
 
 import typer
-from spacy.tokens import Doc, DocBin, SpanGroup
+from spacy.tokens import Doc, DocBin, Span, SpanGroup
 from spacy.training.converters import conll_ner_to_docs
 from wasabi import msg
 
@@ -46,8 +46,11 @@ def convert_iob_to_docs(
     docs_with_spans: List[Doc] = []
 
     for docs in zip(*docs_per_level):
-        spans = [ent for doc in docs for ent in doc.ents]
         doc = docs[0]
+        # recreate all spans for the same underlying doc
+        spans = []
+        for span in [ent for doc in docs for ent in doc.ents]:
+            spans.append(Span(doc, span.start, span.end, span.label_))
         group = SpanGroup(doc, name=spans_key, spans=spans)
         doc.spans[spans_key] = group
         docs_with_spans.append(doc)

experimental/ner_spancat_compare/scripts/transfer_ent_component.py

@@ -3,7 +3,7 @@
 from spacy.tokens.span_group import SpanGroup
 
 
-@Language.factory("transfer-ent.v1", requires=["doc._.ents"])
+@Language.factory("transfer-ent", requires=["doc._.ents"])
 def make_transfer_component(nlp: Language, name: str, span_key: str):
     return TransferEntComponent(nlp, name, span_key)