Update README.md

README.md

@@ -1,22 +1,25 @@
----
-license: mit
-dataset_info:
-  features:
-  - name: query_id
-    dtype: int64
-  - name: document_ids
-    sequence: string
-  - name: scores
-    sequence: float64
-  splits:
-  - name: train
-    num_bytes: 17939263
-    num_examples: 28945
-  download_size: 9649898
-  dataset_size: 17939263
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
----
+---
+license: mit
+dataset_info:
+  features:
+  - name: query_id
+    dtype: int64
+  - name: document_ids
+    sequence: string
+  - name: scores
+    sequence: float64
+  splits:
+  - name: train
+    num_bytes: 17939263
+    num_examples: 28945
+  download_size: 9649898
+  dataset_size: 17939263
+configs:
+- config_name: default
+  data_files:
+  - split: train
+    path: data/train-*
+---
+This is a 29k selection from [500k query/doc/score](https://huggingface.co/datasets/akhooli/ar_mmarco_500k_qs) using JinaAI multilingual reranker. The criteria is  
+stds >= 0.13 and (maxmins >=0.6) and (means > 0.45 and means < 0.95). Documents are in the 
+[Ar_MMARCO](https://huggingface.co/datasets/akhooli/ar_mmarco_docs) dataset (translated)