From cf5b46b63e91b9a2881c3a7d52bb9d2856c809f2 Mon Sep 17 00:00:00 2001
From: Adriane Boyd <adrianeboyd@gmail.com>
Date: Mon, 21 Feb 2022 10:22:36 +0100
Subject: [PATCH] Fix Tok2Vec for empty batches (#10324)

* Add test for tok2vec with vectors and empty docs

* Add shortcut for empty batch in Tok2Vec.predict

* Avoid types
---
 spacy/pipeline/tok2vec.py            |  4 ++++
 spacy/tests/pipeline/test_tok2vec.py | 23 +++++++++++++++++++++--
 2 files changed, 25 insertions(+), 2 deletions(-)

diff --git a/spacy/pipeline/tok2vec.py b/spacy/pipeline/tok2vec.py
index cb601e5dc..2e3dde3cb 100644
--- a/spacy/pipeline/tok2vec.py
+++ b/spacy/pipeline/tok2vec.py
@@ -118,6 +118,10 @@ class Tok2Vec(TrainablePipe):
 
         DOCS: https://spacy.io/api/tok2vec#predict
         """
+        if not any(len(doc) for doc in docs):
+            # Handle cases where there are no tokens in any docs.
+            width = self.model.get_dim("nO")
+            return [self.model.ops.alloc((0, width)) for doc in docs]
         tokvecs = self.model.predict(docs)
         batch_id = Tok2VecListener.get_batch_id(docs)
         for listener in self.listeners:
diff --git a/spacy/tests/pipeline/test_tok2vec.py b/spacy/tests/pipeline/test_tok2vec.py
index eeea906bb..a5ac85e1e 100644
--- a/spacy/tests/pipeline/test_tok2vec.py
+++ b/spacy/tests/pipeline/test_tok2vec.py
@@ -11,7 +11,7 @@ from spacy.lang.en import English
 from thinc.api import Config, get_current_ops
 from numpy.testing import assert_array_equal
 
-from ..util import get_batch, make_tempdir
+from ..util import get_batch, make_tempdir, add_vecs_to_vocab
 
 
 def test_empty_doc():
@@ -140,9 +140,25 @@ TRAIN_DATA = [
 ]
 
 
-def test_tok2vec_listener():
+@pytest.mark.parametrize("with_vectors", (False, True))
+def test_tok2vec_listener(with_vectors):
     orig_config = Config().from_str(cfg_string)
+    orig_config["components"]["tok2vec"]["model"]["embed"][
+        "include_static_vectors"
+    ] = with_vectors
     nlp = util.load_model_from_config(orig_config, auto_fill=True, validate=True)
+
+    if with_vectors:
+        ops = get_current_ops()
+        vectors = [
+            ("apple", ops.asarray([1, 2, 3])),
+            ("orange", ops.asarray([-1, -2, -3])),
+            ("and", ops.asarray([-1, -1, -1])),
+            ("juice", ops.asarray([5, 5, 10])),
+            ("pie", ops.asarray([7, 6.3, 8.9])),
+        ]
+        add_vecs_to_vocab(nlp.vocab, vectors)
+
     assert nlp.pipe_names == ["tok2vec", "tagger"]
     tagger = nlp.get_pipe("tagger")
     tok2vec = nlp.get_pipe("tok2vec")
@@ -169,6 +185,9 @@ def test_tok2vec_listener():
     ops = get_current_ops()
     assert_array_equal(ops.to_numpy(doc.tensor), ops.to_numpy(doc_tensor))
 
+    # test with empty doc
+    doc = nlp("")
+
     # TODO: should this warn or error?
     nlp.select_pipes(disable="tok2vec")
     assert nlp.pipe_names == ["tagger"]