TRECRun is a library for working with TREC run files, with an API heavily inspired by PyTerrier's pipeline operators.
| API | Operator | Description |
|---|---|---|
TRECRun(results) |
Create a TRECRun object from a dictionary of results or a path to a run file in TREC format. |
|
add(self, other), subtract, multiply, divide |
+, -, *, / |
Perform the given operation between self's document scores and other, which can be a TRECRun or a scalar. |
topk(self, k) |
% |
Retain only the top-k documents for each qid after sorting by score. |
intersect(self, other) |
& |
Retain only the queries and documents that appear in both self and other. |
concat(self, other) |
Concat the documents in other and self, with those in other appearing at the end. Their scores will be modified to accomplish this. |
|
normalize(self, method='rr') |
Normalize scores in self using RRF (rr), sklearn's min-max scaling (minmax), or sklearn's scaling (standard). |
|
write_trec_run(self, outf) |
Write self to outfn in TREC format. |
|
evaluate(self, qrels, metrics, return_average=True) |
Compute metrics for self using qrels and return either the average metric or a dict mapping metric names to their values for each QID. |