Fix inconsistency between best_path, best score and scores #949
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Hello @mauryaland thanks for the PR! Could you paste a minimal code example for instance with an example sentence and one of the public NER models to help me better understand the incosistency error? |
|
Hello @alanakbik, here is an example: from flair.data import Sentence
from flair.models SequenceTagger
tagger = SequenceTagger.load('fr-ner')
sentence = Sentence("La France est un beau pays à parcourir.")
tagger.predict(sentence)
print(sentence.get_spans('ner'))
>>> [<LOC-span (1,2): "La France">]
for token in sentence:
print(f"{token} {token.get_tag('ner')} {token.tags_proba_dist} \n\n")
>>> Token: 1 La I-LOC (0.7660126090049744) {'ner': [<unk> (0.0008465868304483593), O (0.7660126090049744), I-PER (0.010933118872344494), I-LOC (0.20101521909236908), I-ORG (0.004218324087560177), I-MISC (0.010774117894470692), B-PER (0.00022157360217534006), B-LOC (0.0013388736406341195), B-MISC (0.0003332003252580762), B-ORG (0.000203109928406775), <START> (0.0), <STOP> (0.004103234503418207)]}
Token: 2 France I-LOC (0.9984954595565796) {'ner': [<unk> (9.132213563134428e-08), O (8.645388849259916e-10), I-PER (0.0001087988493964076), I-LOC (0.9984954595565796), I-ORG (0.0005726731033064425), I-MISC (0.0007387555087916553), B-PER (2.118784436788701e-07), B-LOC (5.522951323655434e-05), B-MISC (6.367911709048713e-08), B-ORG (9.927322963676488e-08), <START> (0.0), <STOP> (2.8619269869523123e-05)]}
....As we can see, using the method to get scores of every tag, the best score is assigned to the tag "O" (from variables scores in the source code). However, the best_score (variable from the code source) is coupled with the best_path, in this case "I-LOC". I hope that my example is clear. It is happening with some models we trained on our own tags too. |
|
@mauryaland thanks for adding the example! |
|
👍 |
1 similar comment
|
👍 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Fixing inconsistency between the couple (best_path, best_scores) and scores in the SequenceTagger model using the CRF for predictions.
Indeed, what is happening sometimes is that the tag with the highest score is not the same between scores and the pair (best_path, best_scores). This PR should fix this behavior.