Upload script/process_from_parquet.py with huggingface_hub

script/process_from_parquet.py

@@ -0,0 +1,31 @@
+
+import dask.dataframe as dd
+from functools import partial
+
+
+def read_parquet_file(parquet_file_path, npartitions=50, top=None):
+
+    print(f"Process parquet file from {parquet_file_path}")
+    file_name = parquet_file_path.split("/")[-1]
+    parquet_df = dd.read_parquet(parquet_file_path, engine="pyarrow")
+    parquet_df = parquet_df.repartition(npartitions=npartitions)  # Smaller partitions
+
+    if top:
+        parquet_df = parquet_df.head(top, compute=False) # compute=False to keep it as DaskDataframe
+
+    return parquet_df, file_name
+
+
+def process_parquet_df(parquet_df, file_name, process_row_func, process_partition):
+
+    # A new function of process_row_func to allow pre-defining parameters of process-row_func. 
+    process_row_with_params = partial(process_row_func, parquet_file_name=file_name)
+    result_df = parquet_df.map_partitions(process_partition, process_row_with_params)
+
+    return result_df
+
+
+def save_to_csv(df, final_path):
+    # Save the processed DataFrame to csv
+    df.to_csv(final_path, index=False, single_file=True)
+