Created
September 7, 2023 13:11
-
-
Save WojciechKusa/e8614f262f8aa72735b6cafa0667ec6b to your computer and use it in GitHub Desktop.
Convert WSS score to TNR
This file contains 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 typing import Dict, List, Union | |
import numpy as np | |
ScoreType = Union[float, List[float], Dict[str, float]] | |
def convert_single_score(score: float, min_score: float, max_score: float) -> float: | |
""" | |
Convert a single score using min-max normalization. | |
Args: | |
score (float): Score to be converted. Score should not be larger than 1. | |
min_score (float): Minimum score value. | |
max_score (float): Maximum score value. | |
Returns: | |
float: Converted score. | |
""" | |
if score > 1: | |
raise ValueError( | |
f"Invalid score value {score}. Score values should not be larger than 1." | |
) | |
return (score - min_score) / (max_score - min_score) | |
def wss_to_tnr( | |
scores: ScoreType, | |
dataset_size: int, | |
num_relevant: int, | |
recall: float = 0.95, | |
) -> ScoreType: | |
""" | |
Convert WSS@95% scores to TNR@95% scores using min-max normalization. | |
Args: | |
scores (Union[float, List[Union[float, str]], Dict[str, float]]): WSS@95% scores to be converted. | |
WSS scores should be between 0 and 1. | |
dataset_size (int): Total size of the dataset. | |
num_relevant (int): Number of relevant documents in the dataset. | |
recall (float): Recall value used in the conversion. | |
Returns: | |
Union[float, List[float], Dict[str, float]]: Converted TNR@95% scores. | |
Formula: | |
TNR@95% = (WSS@95 - min(WSS@95)) / (max(WSS@95) - min(WSS@95)). | |
max(WSS@95) = (num_irrelevant + 0.05 * num_relevant) / dataset_size - 0.05 | |
min(WSS@95) = (0.05 * num_relevant) / dataset_size - 0.05 | |
""" | |
num_irrelevant = dataset_size - num_relevant | |
max_score = (num_irrelevant + (1 - recall) * num_relevant) / dataset_size - ( | |
1 - recall | |
) | |
min_score = ((1 - recall) * num_relevant) / dataset_size - (1 - recall) | |
if isinstance(scores, float): | |
return convert_single_score(scores, min_score, max_score) | |
if isinstance(scores, list): | |
try: | |
return [ | |
convert_single_score(score, min_score, max_score) for score in scores | |
] | |
except ValueError: | |
raise ValueError("The list contains invalid non-numeric data.") | |
if isinstance(scores, dict): | |
return { | |
model: convert_single_score(score, min_score, max_score) | |
for model, score in scores.items() | |
} | |
raise TypeError( | |
"scores should be either a float, a list of floats, or a dictionary with model names as keys and scores as values." | |
) | |
if __name__ == "__main__": | |
dataset_size = 2544 | |
num_relevant = 41 | |
wss = 0.566 | |
expected_tnr = 0.625 | |
tnr = wss_to_tnr(wss, dataset_size, num_relevant) | |
print(tnr) | |
assert np.isclose(tnr, expected_tnr, atol=0.001) | |
scores = [0.566, 0.523, 0.733, 0.801, 0.787, 0.783, 0.783] | |
expected_tnrs = [0.625, 0.582, 0.795, 0.864, 0.850, 0.846, 0.846] | |
tnrs = wss_to_tnr(scores, dataset_size, num_relevant) | |
print(tnrs) | |
assert np.isclose(tnrs, expected_tnrs, atol=0.001).all() |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment