Spaces:
Sleeping
Sleeping
File size: 4,477 Bytes
d891c3c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 |
import datasets
import evaluate
import numpy as np
_CITATION = """\
@inproceedings{palotti2019,
author = {Palotti, Joao and Scells, Harrisen and Zuccon, Guido},
title = {TrecTools: an open-source Python library for Information Retrieval practitioners involved in TREC-like campaigns},
series = {SIGIR'19},
year = {2019},
location = {Paris, France},
publisher = {ACM}
}
"""
_DESCRIPTION = """\
A metric to evaluate ranking tasks using the TREC evaluation tool. It compares predicted rankings of items (e.g., documents) with their true relevance scores. The metric takes two inputs: references (true relevance scores) and predictions (predicted scores), both as lists of lists, where each (i, j) is the truth or the predicted score of the document j in the query i. In a nutshell: simplifies the usage of TREC to compute ranking metrics given scores per sample.
"""
_KWARGS_DESCRIPTION = """ Computes MAP, P@K, RR, and NDCG using the TREC evaluation tool.
Args:
references (list(list(float))): true scores for each query
predictions (list(list(float))): pred scores for each query
Returns:
Dict: the set of TREC's metrics scores
Example:
# (i, j) means the truth/predicted score of the document j in the query i
references = [[5, 0, 3, 0, 0, 2, 1],
[5, 0, 3, 0, 0, 2, 1],
[5, 0, 3, 0, 0, 2, 1],
[0, 1, 2]]
predictions = [[3, 4, 2, 0, 1, 5, 0],
[2, 0, 4, 5, 0, 1, 3],
[0, 3, 2, 1, 5, 0, 4],
[5, 3, 2]]
metric = evaluate.load("symanto/ranking_evaluator")
metric.compute(references=references, predictions=predictions)
"""
class RankingEvaluator(evaluate.Metric):
def _info(self):
return evaluate.MetricInfo(
description=_DESCRIPTION,
citation=_CITATION,
inputs_description=_KWARGS_DESCRIPTION,
features=datasets.Features(
{
"predictions": datasets.Sequence(datasets.Value("float32")),
"references": datasets.Sequence(datasets.Value("float32")),
}
),
)
def _download_and_prepare(self, dl_manager):
self.trec_eval = evaluate.load("trec_eval")
def _compute(
self, references: list[list[float]], predictions: list[list[float]]
) -> dict:
"""
Calculates MAP, P@K, RR, and NDCG using the TREC evaluation tool.
Args:
references (list(list(float))): true scores for each query
predictions (list(list(float))): pred scores for each query
Returns:
Dict: the set of TREC's metrics scores
Example:
# (i, j) means the truth/predicted score of the document j in the query i
references = [[5, 0, 3, 0, 0, 2, 1],
[5, 0, 3, 0, 0, 2, 1],
[5, 0, 3, 0, 0, 2, 1],
[0, 1, 2]]
predictions = [[3, 4, 2, 0, 1, 5, 0],
[2, 0, 4, 5, 0, 1, 3],
[0, 3, 2, 1, 5, 0, 4],
[5, 3, 2]]
metric = evaluate.load("symanto/ranking_evaluator")
metric.compute(references=references, predictions=predictions)
"""
qrel = {}
run = {}
# Fill qrel
for query_idx, truth in enumerate(references):
for item_idx, relevance in enumerate(truth):
if relevance > 0:
qrel.setdefault("query", []).append(query_idx)
qrel.setdefault("q0", []).append("q0")
qrel.setdefault("docid", []).append(f"doc_{item_idx}")
qrel.setdefault("rel", []).append(relevance)
# Fill run
for query_idx, pred in enumerate(predictions):
ranking = np.argsort(np.argsort(pred)[::-1])
for item_idx, score in enumerate(pred):
if score > 0:
run.setdefault("query", []).append(query_idx)
run.setdefault("q0", []).append("q0")
run.setdefault("docid", []).append(f"doc_{item_idx}")
run.setdefault("score", []).append(score)
run.setdefault("system", []).append("sys")
run.setdefault("rank", []).append(ranking[item_idx])
return self.trec_eval.compute(references=[qrel], predictions=[run])
|