Skip to content

Fix error on large documents #4

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

Open
edwintorok wants to merge 1 commit into
base: master
Choose a base branch
from
Open

Fix error on large documents #4

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
33 changes: 17 additions & 16 deletions qanom/spacy_component_nominalization_detector.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,32 +40,33 @@ def __call__(self, doc: Doc) -> Doc:
token._.verb_form = verb_form
# Apply classifier at sentence level
# Prepare Tokenization
tokenized = self.tokenizer(doc.text, return_tensors="pt", padding=True).to(self.device)
# get mapping of word to tokens (for using indexes from dataset)
wordId2numOfTokens = [len(self.tokenizer.tokenize(t.text)) for t in doc]
wordId2firstTokenId, curr_tok_id = [], 1
for wordId in range(len(wordId2numOfTokens)):
wordId2firstTokenId.append(curr_tok_id)
curr_tok_id += wordId2numOfTokens[wordId]
tokenized = self.tokenizer([sent.text for sent in doc.sents], return_tensors="pt", padding=True).to(self.device)
# Run binary classification model on candidate nouns
model = self.candidate_classifier.to(self.device)
logits = model(tokenized.input_ids).logits
# compose binary probability as the mean of positive label's prob with complement of neg. label's prob.
positive_lbl_id = 0 # because label_map[0] == 'True'
negative_lbl_id = 1 # because label_map[1] == 'False'
positive_probability = logits[0,:,positive_lbl_id].sigmoid() # probability of the "true" label (index 0 in axis-2)
negative_probability = logits[0,:,negative_lbl_id].sigmoid()
positive_probability = logits[:,:,positive_lbl_id].sigmoid() # probability of the "true" label (index 0 in axis-2)
negative_probability = logits[:,:,negative_lbl_id].sigmoid()
probability = (positive_probability + (1-negative_probability) ) / 2
preds = probability > self.threshold
# iterate over tokens and set `is_nominalization` and `is_nominalization_confidence` attributes for candidate nouns
nominalizations = []
for token in doc:
if token._.is_candidate_nominalization:
token_id = wordId2firstTokenId[token.i]
token._.is_nominalization = preds[token_id].item()
token._.is_nominalization_confidence = probability[token_id].item()
if token._.is_nominalization:
nominalizations.append(token)
for i, sent in enumerate(doc.sents):
# get mapping of word to tokens (for using indexes from dataset)
wordId2numOfTokens = [len(self.tokenizer.tokenize(t.text)) for t in sent]
wordId2firstTokenId, curr_tok_id = [], 1
for wordId in range(len(wordId2numOfTokens)):
wordId2firstTokenId.append(curr_tok_id)
curr_tok_id += wordId2numOfTokens[wordId]
for token_i, token in enumerate(sent):
if token._.is_candidate_nominalization:
token_id = wordId2firstTokenId[token_i]
token._.is_nominalization = preds[i][token_id].item()
token._.is_nominalization_confidence = probability[i][token_id].item()
if token._.is_nominalization:
nominalizations.append(token)

# Add `nominalizations` attribute to doc
doc._.nominalizations = nominalizations
Expand Down