Add dataset conversion script for MLSUM

convert_datasets_to_json.py

@@ -0,0 +1,39 @@
+from datasets import (get_dataset_config_names, get_dataset_split_names,
+                      load_dataset)
+from huggingface_hub import list_datasets
+
+
+def convert(dataset_id: str):
+    dataset_name = dataset_id.split("/")[-1]
+    configs = get_dataset_config_names(dataset_id)
+
+    for config in configs:
+        splits = get_dataset_split_names(dataset_id, config)
+        splits = [split for split in splits if split not in ["train", "validation"]]
+        for split in splits:
+            columns_to_keep = ["gem_id", "gem_parent_id", "target"]
+            dataset = load_dataset(dataset_id, name=config, split=split)
+            # It seems like we store the references column as the target one
+            dataset = dataset.map(lambda x: {"target": x["references"]})
+            # Delete unused columns
+            # The test split doesn't have a parent ID
+            if split == "test":
+                columns_to_keep.remove("gem_parent_id")
+            # The `datasets` JSON serializer is buggy - use `pandas` for now
+            df = dataset.to_pandas()
+            df[columns_to_keep].to_json(f"{dataset_name}_{config}_{split}.json", orient="records")
+            # TODO: validate against existing references on GitHub
+            # diff <(jq --sort-keys . mlsum_de_challenge_test_covid.json) <(jq --sort-keys . ~/git/GEM-metrics/data/references/mlsum_de_challenge_test_covid.json)
+
+
+def main():
+    all_datasets = list_datasets()
+    gem_datasets = [dataset for dataset in all_datasets if dataset.id.startswith("GEM/")]
+    # Test run with MLSUM
+    mlsum_datasets = [dataset for dataset in gem_datasets if dataset.id.startswith("GEM/mlsum")]
+    for dataset in mlsum_datasets:
+        convert(dataset.id)
+
+
+if __name__ == "__main__":
+    main()