Last active
August 29, 2025 00:19
-
-
Save myedibleenso/c065db1a9d293a425c46bd031b2aab3b to your computer and use it in GitHub Desktop.
Sketch for Seetha. "fuzzy" alignment to ground truth for taxonomy
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
| from __future__ import annotations | |
| from collections import Counter | |
| from typing import Dict, List, Optional, Text | |
| from dataclasses import dataclass | |
| @dataclass | |
| class Span: | |
| start: int | |
| end: int | |
| tokens: Optional[List[Text]] | |
| # the Gold label from the taxonomy | |
| label: Optional[Text] | |
| def __len__(self) -> int: | |
| # +1 ? | |
| return self.end - self.start | |
| def overlaps(self, other: Span) -> bool: | |
| ## compare start and end of two spans to determine if they overlap | |
| pass | |
| def subsumes(self, other: Span) -> bool: | |
| """ | |
| Determine if other Span is included in this Span. | |
| """ | |
| def __eq__(self, other: Span) -> bool: | |
| return (self.start == other.start and self.end == other.end) | |
| def score_match(self, other: Span) -> float: | |
| """ | |
| Returns a score between 0 and 1 where 1 is a perfect match. | |
| """ | |
| # is spans are equal, it's a perfect match | |
| if self == other: | |
| return 1.0 | |
| elif self.subsumes(other): | |
| # what percentage of our span is matched? | |
| return len(other) / len(self) | |
| elif self.overlaps(other): | |
| # how much do they intersect? | |
| # how long is each? | |
| # FIXME: generate a score here | |
| return 0.0 | |
| # tests | |
| # assuming an **exclusive** interval | |
| s_a = Span(start=0, end=3) | |
| assert len(s_a) == 3 | |
| # FIXME: add simple tests for .overlaps(), ==, etc. | |
| predicted = Span(start=0, end=3) | |
| # gold spans | |
| gold_candidates = [Span(start=0, end=1), Span(start=2, end=5)] | |
| scored_pairs = [(predicted.score_match(candidate) ,candidate) for candidate in gold_candidates] | |
| top_score, _ = max(scored_pairs) | |
| # scenario 1: find all gold spans with our highest score and create **multiple** training examples (i.e., one example for each **distinct** label) | |
| # scenario 2: find all gold spans with our highest score and create a training example for our most frequent label with that score. Note that in the case of a tie, this may produce multiple examples for one original prediction. | |
| highest_scoring: Dict[Text, int] = Counter([candidate.label for (score, candidate) in scored_pairs if score == top_score]) | |
| _, most_freq = highest_scoring.most_common(1) | |
| # all labels with our highest frequency | |
| to_assign = [lbl for (lbl, cnt) in highest_scoring.most_common() if cnt == most_freq] | |
| for lbl in to_assign: | |
| # TODO: process our prediction by writing the CoNLL-style sentence with the new label? | |
| # for example, a two column tsv file: | |
| # I like <span>turtles</span> . LABEL_X | |
| pass |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment