fix neg:nil issie while loading dataset

README.md

@@ -30,14 +30,14 @@ dataset_info:
       dtype: float
   splits:
     - name: train
-      num_bytes: 1
-      num_examples: 1
+      num_bytes: 89609915
+      num_examples: 502939
     - name: test
-      num_bytes: 1
-      num_examples: 1      
+      num_bytes: 969945
+      num_examples: 43
     - name: dev
-      num_bytes: 1
-      num_examples: 1      
+      num_bytes: 1206403
+      num_examples: 6980
 train-eval-index:
   - config: default
     task: sentence-similarity

convert.py

@@ -67,9 +67,7 @@ def process(
             for rel in rels
             if rel.doc < len(corpus) and rel.score == 0 and corpus[rel.doc] != ""
         ]
-        group = {"query": queries[query], "pos": pos}
-        if len(neg) > 0:
-            group["neg"] = neg
+        group = {"query": queries[query], "pos": pos, "neg": neg}
         result.append(group)
     return result
 
@@ -78,17 +76,18 @@ def main():
     parser = HfArgumentParser((ConversionAgruments))
     (args,) = parser.parse_args_into_dataclasses()
     print(f"Args: {args}")
-    corpus = load_json(f"{args.path}/corpus.jsonl", split="train")
-    queries = load_json(f"{args.path}/queries.jsonl")
     qrels = {
-        "dev": process(load_qrel(f"{args.path}/qrels/dev.tsv"), queries, corpus),
-        "test": process(load_qrel(f"{args.path}/qrels/test.tsv"), queries, corpus),
-        "train": process(load_qrel(f"{args.path}/qrels/train.tsv"), queries, corpus),
+        "dev": load_qrel(f"{args.path}/qrels/dev.tsv"),
+        "train": load_qrel(f"{args.path}/qrels/train.tsv"),
+        "test": load_qrel(f"{args.path}/qrels/test.tsv"),
     }
+    corpus = load_json(f"{args.path}/corpus.jsonl", split="train")
+    queries = load_json(f"{args.path}/queries.jsonl")
     print("processing done")
     for split, data in qrels.items():
+        dataset = process(data, queries, corpus)
         with open(f"{args.out}/{split}.jsonl", "w") as out:
-            for item in data:
+            for item in dataset:
                 json.dump(item, out)
                 out.write("\n")
     print("done")

data/dev/dev.jsonl.gz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:92ff05e37188b132fdac1734e823e78b8c92a2e00622f9886711b0d7d9274624
-size 1109843
+oid sha256:dbd532e51439da3bd74c79105f9f9f27beafd1db73a1b2155713ec37c8fd376d
+size 1206403

data/test/test.jsonl.gz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3c4362a2f021d6b45f068d2f13aee9336b84123e718df797930956a540f30167
-size 860084
+oid sha256:1b454c2d3e35adf12716e64699cc6637f703d9e82fc46f036cfa5af59433edc5
+size 969945

data/train/train.jsonl.gz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cfe4888522ab3a7be8bfa406f3e45a207193cd8b0fd026acd895cb6c1213a86f
-size 92239408
+oid sha256:d56a52df9c3bbd14ee6f92cef2aeb61bcf6d33bc9e85012559b53a5f246157a0
+size 89609915

requirements.txt

@@ -1 +1,2 @@
 datasets
+transformers

use.py

@@ -0,0 +1,3 @@
+from datasets import load_dataset
+
+data = load_dataset("nixiesearch/MSMARCO", split="train")