-
Notifications
You must be signed in to change notification settings - Fork 0
/
RetrievalEvaluation.py
201 lines (164 loc) · 7.88 KB
/
RetrievalEvaluation.py
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
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
import pickle
from abc import ABC, abstractmethod
from typing import Dict, List
import pandas as pd
from trectools import TrecEval, TrecQrel, TrecRun
def _predictions_dict_to_df(query_id_to_pmids: Dict[int, List[int]]) -> pd.DataFrame:
predictions = {
"query": [],
"q0": [],
"docid": [],
"rank": [],
"score": [],
"system": [],
}
for query_id, pmids in query_id_to_pmids.items():
n = len(pmids)
predictions["query"].extend([query_id] * n)
predictions["q0"].extend(["0"] * n)
predictions["docid"].extend(pmids)
predictions["rank"].extend(list(range(n)))
predictions["score"].extend([0] * n)
predictions["system"].extend(["system"] * n)
return pd.DataFrame.from_dict(predictions)
def _trectools_object_to_dict(trec_eval: TrecEval) -> Dict[str, float]:
result = {}
result["num_ret"] = trec_eval.get_retrieved_documents(per_query=False)
result["num_rel"] = trec_eval.get_relevant_documents(per_query=False)
result["num_rel_ret"] = trec_eval.get_relevant_retrieved_documents(per_query=False)
result["num_q"] = len(trec_eval.run.topics())
result["map"] = trec_eval.get_map(depth=10000, per_query=False, trec_eval=True)
result["gm_map"] = trec_eval.get_geometric_map(depth=10000, trec_eval=True)
result["bpref"] = trec_eval.get_bpref(depth=1000, per_query=False, trec_eval=True)
result["Rprec"] = trec_eval.get_rprec(depth=1000, per_query=False, trec_eval=True)
result["recip_rank"] = trec_eval.get_reciprocal_rank(depth=1000, per_query=False, trec_eval=True)
for v in [5, 10, 15, 20, 30, 100, 200, 500, 1000]:
result[f"P@{v}"] = trec_eval.get_precision(depth=v, per_query=False, trec_eval=True)
for v in [5, 10, 15, 20, 30, 100, 200, 500, 1000]:
result[f"NDCG@{v}"] = trec_eval.get_ndcg(depth=v, per_query=False, trec_eval=True)
return result
class RetrievalEvaluator(ABC):
@abstractmethod
def get_query_id_to_query(self) -> Dict[int, str]:
"""
Returns the queries to use for the evaluation. Each query has a unique query ID (integer).
:return: A dictionary from query ID to query text.
"""
pass
@abstractmethod
def evaluate(self, query_id_to_pmids: Dict[int, List[int]]) -> Dict[str, float]:
"""
Evaluate performance of the retrieval system, computing a large range of evaluation metrics.
:param query_id_to_pmids: A dictionary from query ID to a ranked list of retrieved PMIDs.
:return: A dictionary from metric name to metric value. See
https://people.cs.georgetown.edu/~nazli/classes/ir-Slides/Evaluation-12.pdf for a nice overview of the various
metrics.
"""
pass
class TrecCovidEvaluator(RetrievalEvaluator):
"""
Evaluation using subset of the TREC COVID set (https://ir.nist.gov/trec-covid/), limited to those documents that are
in PubMed.
During TREC, the best observed performance for an automatic run was:
P@5: 0.7800
NDCG@10: 0.6080
MAP: 0.3128
bpref: 0.4832
But keep in mind this score was achieved before the set was publicly available. Future retrieval approaches will
have been trained or fine-tuned on this set, and may show overly-optimistic results.
"""
def __init__(self):
with open("TREC_COVID.pickle", "rb") as f:
dataset = pickle.load(f)
self.query_id_to_qrels = dataset["query_id_to_qrels"]
self.query_id_to_query = dataset["query_id_to_query"]
self.allowed_pmids = set(dataset["pmids"])
def get_query_id_to_query(self) -> Dict[int, str]:
return self.query_id_to_query
def _filter_pmids(self, query_id, pmids):
return [pmid for pmid in pmids if pmid in self.allowed_pmids]
def evaluate(self, query_id_to_pmids: Dict[int, List[int]]) -> Dict[str, float]:
references = {
"query": [],
"q0": [],
"docid": [],
"rel": []
}
for query_id, pmid_to_score in self.query_id_to_qrels.items():
n = len(pmid_to_score)
references["query"].extend([query_id] * n)
references["q0"].extend(["0"] * n)
references["docid"].extend(pmid_to_score)
references["rel"].extend(pmid_to_score.values())
query_id_to_pmids = {query_id: self._filter_pmids(query_id, pmids) for query_id, pmids in query_id_to_pmids.items()}
df_run = _predictions_dict_to_df(query_id_to_pmids)
df_qrel = pd.DataFrame.from_dict(references)
trec_run = TrecRun()
trec_run.filename = "placeholder.file"
trec_run.run_data = df_run
trec_qrel = TrecQrel()
trec_qrel.filename = "placeholder.file"
trec_qrel.qrels_data = df_qrel
trec_eval = TrecEval(trec_run, trec_qrel)
return _trectools_object_to_dict(trec_eval)
class BioASQTrain2024Evaluator(RetrievalEvaluator):
"""
Evaluation using the BioASQ Task B training set of 2024
(http://participants-area.bioasq.org/general_information/Task12b/)
"""
def __init__(self, use_sample: bool = False):
with open("BioASQTrain2024.pickle", "rb") as f:
dataset = pickle.load(f)
self.query_id_to_query = dataset["query_id_to_query"]
self.query_id_to_relevant_pmids = dataset["query_id_to_pmids"]
if use_sample:
sampled_query_ids = dataset["sampled_query_ids"]
self.query_id_to_query = {query_id: self.query_id_to_query[query_id] for query_id in sampled_query_ids}
self.query_id_to_relevant_pmids = {query_id: self.query_id_to_relevant_pmids[query_id] for query_id in sampled_query_ids}
self.max_pmid = dataset["max_pmid"]
def get_query_id_to_query(self) -> Dict[int, str]:
return self.query_id_to_query
def _filter_pmids(self, pmids: List[int]) -> List[int]:
return [pmid for pmid in pmids if pmid <= self.max_pmid]
def evaluate(self, query_id_to_pmids: Dict[int, List[int]]) -> Dict[str, float]:
references = {
"query": [],
"q0": [],
"docid": [],
"rel": []
}
for query_id, relevant_pmids in self.query_id_to_relevant_pmids.items():
pmids = query_id_to_pmids[query_id]
non_relevant_pmids = [pmid for pmid in pmids if pmid not in relevant_pmids]
n = len(relevant_pmids) + len(non_relevant_pmids)
references["query"].extend([query_id] * n)
references["q0"].extend(["0"] * n)
references["docid"].extend(relevant_pmids)
references["docid"].extend(non_relevant_pmids)
references["rel"].extend([1]*len(relevant_pmids))
references["rel"].extend([0] * len(non_relevant_pmids))
df_qrel = pd.DataFrame.from_dict(references)
query_id_to_pmids = {query_id: self._filter_pmids(pmids) for query_id, pmids in query_id_to_pmids.items()}
df_run = _predictions_dict_to_df(query_id_to_pmids)
trec_run = TrecRun()
trec_run.filename = "placeholder.file"
trec_run.run_data = df_run
trec_qrel = TrecQrel()
trec_qrel.filename = "placeholder.file"
trec_qrel.qrels_data = df_qrel
trec_eval = TrecEval(trec_run, trec_qrel)
return _trectools_object_to_dict(trec_eval)
if __name__ == "__main__":
# with open("TREC_COVID.pickle", "rb") as f:
# dataset = pickle.load(f)
# query_id_to_qrels = dataset["query_id_to_qrels"]
# query_id_to_pmids = {query_id: pmid_to_score.keys() for query_id, pmid_to_score in query_id_to_qrels.items()}
# evaluator = TrecCovidEvaluator()
# results = evaluator.evaluate(query_id_to_pmids)
# print(results)
with open("BioASQTrain2024.pickle", "rb") as f:
dataset = pickle.load(f)
query_id_to_pmids = dataset["query_id_to_pmids"]
evaluator = BioASQTrain2024Evaluator()
results = evaluator.evaluate(query_id_to_pmids)
print(results)