Try to download manually

app.py

@@ -1,10 +1,15 @@
+import os
+from pathlib import Path
+from urllib.request import urlretrieve
+
 import cartopy.crs as ccrs
 import fugue.api as fa
 import geopandas as gpd
 import geoviews as gv
 import panel as pn
+import pandas as pd
 import pyarrow as pa
-from datasets import load_dataset, Split
+from datasets import get_dataset_infos
 from holoviews.streams import RangeXY
 from shapely import wkt
 
@@ -34,17 +39,31 @@ INTRO = """
     and all their supporting dependencies.
 """
 
+DATA_DIR = Path.home() / ".cache" / "huggingface" / "datasets"
+DATA_PATH = DATA_DIR / "edges.parquet"
+
 QUERY_FMT = """
-    df = CREATE USING load_hf(path="ahuang11/tiger_layer_edges")
+    df = LOAD_DATASET({{path}})
     df_sel = SELECT STATEFP, COUNTYFP, FULLNAME, geometry \
         FROM df WHERE FULLNAME == '{{name}}'
 """
 
 
-def load_hf(path:str) -> pa.Table:
-    dataset = load_dataset(path, split=Split.ALL).data
-    table = pa.Table.from_batches(dataset.to_batches())
-    return table
+def download_hf(path: str, **kwargs):
+    info = list(get_dataset_infos(path).values())[0]
+    file_df = pd.DataFrame({"file": list(info.download_checksums.keys())})
+    DATA_PATH.makedirs(exist_ok=True)
+
+    def _download(df: pd.DataFrame) -> None:
+        for file in df["file"]:
+            file_name = os.path.basename(file)
+            data_path = DATA_PATH / file_name
+            done_path = data_path.with_suffix(".done")
+            if not done_path.exists():
+                urlretrieve(file, data_path)
+                done_path.touch()
+
+    fa.out_transform(file_df, _download, **kwargs)
 
 
 class MapnStreets:
@@ -66,6 +85,7 @@ class MapnStreets:
         pn.state.onload(self.onload)
 
     def onload(self):
+        download_hf("ahuang11/tiger_layer_edges")
         self.name_input.param.trigger("value")
 
         range_xy = RangeXY()
@@ -92,14 +112,19 @@ class MapnStreets:
         try:
             name = name.strip()
             self.holoviews_pane.loading = True
-            query_fmt = QUERY_FMT
             if "*" in name or "%" in name:
                 name = name.replace("*", "%")
                 query_fmt = query_fmt.replace("==", "LIKE")
             if name == "%":
                 return
             df = fa.as_pandas(
-                fa.fugue_sql(query_fmt, name=name, engine="duckdb", as_local=True)
+                fa.fugue_sql(
+                    query_fmt,
+                    data_path=str(DATA_PATH.absolute()),
+                    name=name,
+                    engine="duckdb",
+                    as_local=True,
+                )
             )
             self.gdf = self.serialize_geom(df)
             county_gdf = self.gdf.drop_duplicates(