Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Fix speed problem with top_k>1 on CPU in edit tree lemmatizer #12017

Merged
merged 10 commits into from
Jan 20, 2023
Merged

Fix speed problem with top_k>1 on CPU in edit tree lemmatizer #12017

Changes from 1 commit
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
278181a
Refactor _scores2guesses
richardpaulhudson Dec 22, 2022
9430306
Handle arrays on GPU
richardpaulhudson Dec 22, 2022
95a4835
Convert argmax result to raw integer
richardpaulhudson Dec 22, 2022
361b64e
Use NumpyOps() to copy data to CPU
richardpaulhudson Dec 23, 2022
4daf5e9
Changes based on review comments
richardpaulhudson Dec 23, 2022
8d487bf
Use different _scores2guesses depending on tree_k
richardpaulhudson Jan 10, 2023
781661b
Add tests for corner cases
richardpaulhudson Jan 10, 2023
00b1d9a
Add empty line for consistency
richardpaulhudson Jan 10, 2023
175b487
Improve naming
richardpaulhudson Jan 17, 2023
dd5975a
Improve naming
richardpaulhudson Jan 17, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
23 changes: 7 additions & 16 deletions spacy/pipeline/edit_tree_lemmatizer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -160,27 +160,18 @@ def predict(self, docs: Iterable[Doc]) -> List[Ints2d]:
def _scores2guesses(self, docs, scores):
guesses = []
for doc, doc_scores in zip(docs, scores):
if self.top_k == 1:
doc_guesses = doc_scores.argmax(axis=1).reshape(-1, 1)
else:
doc_guesses = np.argsort(doc_scores)[..., : -self.top_k - 1 : -1]

if not isinstance(doc_guesses, np.ndarray):
doc_guesses = doc_guesses.get()

doc_compat_guesses = []
for token, candidates in zip(doc, doc_guesses):
tree_id = -1
for candidate in candidates:
for i, token in enumerate(doc):
for _ in range(self.top_k):
richardpaulhudson marked this conversation as resolved.
Show resolved Hide resolved
candidate = doc_scores[i].argmax()
richardpaulhudson marked this conversation as resolved.
Show resolved Hide resolved
candidate_tree_id = self.cfg["labels"][candidate]

if self.trees.apply(candidate_tree_id, token.text) is not None:
tree_id = candidate_tree_id
doc_compat_guesses.append(candidate_tree_id)
break
doc_compat_guesses.append(tree_id)

doc_scores[i, candidate] = -1
richardpaulhudson marked this conversation as resolved.
Show resolved Hide resolved
else:
doc_compat_guesses.append(-1)
guesses.append(np.array(doc_compat_guesses))

return guesses

def set_annotations(self, docs: Iterable[Doc], batch_tree_ids):
Expand Down