changed data type for empty values

prepare_data.py

@@ -7,7 +7,6 @@ import datasets
 from datasets import load_dataset
 
 
-
 def save_and_compress(dataset: Union[datasets.Dataset, pd.DataFrame], name: str, idx=None):
     if idx:
         path = f"{name}_{idx}.jsonl"
@@ -21,8 +20,6 @@ def save_and_compress(dataset: Union[datasets.Dataset, pd.DataFrame], name: str,
     os.system(f'xz -zkf -T0 {path}')  # -TO to use multithreading
 
 
-
-
 def get_dataset_column_from_text_folder(folder_path):
     return load_dataset("text", data_dir=folder_path, sample_by="document", split='train').to_pandas()['text']
 
@@ -56,4 +53,5 @@ for split in ["train", "test"]:
         df['summary/full'] = get_dataset_column_from_text_folder(summary_full_path)
         dfs.append(df)
     df = pd.concat(dfs)
+    df = df.fillna("")  # NaNs can lead to huggingface not recognizing the feature type of the column
     save_and_compress(df, f"data/{split}")

test.jsonl.xz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:313dd03ac209312faa6ed75128d54878949bb8dad3fa8622c6936e6d5231dd41
-size 2323776
+oid sha256:e56dd68c9d505c9c7850f32f9cd883a8e6737585f481c5aff35eb98623449098
+size 2323792

train.jsonl.xz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bc4714244b4cec524bfa56ee718f855d4d33e7a0fa3df595f9ce036ac5ac1ecc
-size 50589476
+oid sha256:058ff61ae9c1c8f25c9bcaf3a33b3f472c74fde69b4aab7039121304e51f8ba5
+size 50582572