How to load the dataset

README.md

@@ -12,11 +12,13 @@ size_categories:
 viewer: false
 configs:
 - config_name: default
-  data_files: "data/wikipedia_shard_0.jsonl.gz"
+  data_files: "data/*.jsonl.gz"
   chunksize: 100
 ---
 # English USAS Mosaico Dataset
 
+*Note* that this dataset cannot be loaded using the HuggingFace datasets library as we have 2 data types within a list in the `pos` column, however you can download the dataset and load it within Python using the `json` and `gzip` core modules, an example loading script can be found at [./example_loading_script.py](./example_loading_script.py).
+
 This dataset contains the processed English Wikipedia pages of the [Mosaico](https://github.com/SapienzaNLP/mosaico/tree/main) dataset, of which this is a subset of the original English Wikipedia dataset whereby it contains only the pages with the tags `good` and `featured`.
 
 Each entry in the dataset is a processed Wikipedia page containing the following annotations/tags:

example_loading_script.py

@@ -0,0 +1,13 @@
+from pathlib import Path
+import gzip
+import json
+
+# Loops over each JSONL file
+for data_file in Path("data").glob("*.jsonl.gz"):
+    print(data_file)
+    # Opens the file
+    with gzip.open(str(data_file), "rt", encoding="utf-8") as f:
+        # Reads the first line of the file and prints it and then breaks
+        for line in f:
+            print(json.loads(line))
+            break