nreimers commited on
Commit
dc4c5a8
1 Parent(s): ab0e9da
Files changed (1) hide show
  1. README.md +2 -3
README.md CHANGED
@@ -72,13 +72,12 @@ The format is the following:
72
  - "top1k_offsets": Passage ID (int) when the numpy matrices are loaded sequentially and vertically stacked
73
  - "top1k_passage_ids": Passage ID (string) as they appear in the dataset
74
  - "top1k_cossim": Cosine similarities
75
- - "qrels": Relevance annotations for the 215 annotated queries by NIST
76
 
77
  ### Queries - JSONL
78
  The folder `queries_jsonl/` contains the queries in a `.jsonl.gz` format.
79
 
80
- Load this .jsonl.gz format is significantly faster than the parquet file.
81
-
82
 
83
  ### Queries - Parquet
84
 
 
72
  - "top1k_offsets": Passage ID (int) when the numpy matrices are loaded sequentially and vertically stacked
73
  - "top1k_passage_ids": Passage ID (string) as they appear in the dataset
74
  - "top1k_cossim": Cosine similarities
75
+ - "qrels": Relevance annotations for the 215 annotated queries by NIST. The **document relevance** scores are provided. You can get the doc_id for a passage via `row['_id'].split("#")[0]`
76
 
77
  ### Queries - JSONL
78
  The folder `queries_jsonl/` contains the queries in a `.jsonl.gz` format.
79
 
80
+ Note: qrels are provided here as a dictionary lookup, while in the parquet format as a list in the format `[doc_id, score]` due to the limited support for dictionaries in parquet.
 
81
 
82
  ### Queries - Parquet
83