Improve arrow_to_dataframe for large amounts of columns (#73)

Co-authored-by: Justin Gilmer <16173348+justinGilmer@users.noreply.github.com>

btrdb/transformers.py

@@ -181,6 +181,19 @@ def arrow_to_dataframe(streamset, agg=None, name_callable=None) -> pd.DataFrame:
     -----
     This method is available for commercial customers with arrow-enabled servers.
     """
+
+    def _rename(col_name, col_names_map):
+        if col_name == "time":
+            return col_name
+        col_name_parts = col_name.split("/")
+        if len(col_name_parts) == 1:
+            uuid = col_name_parts[0]
+            new_col_name = col_names_map[uuid]
+        elif len(col_name_parts) == 2:
+            uuid, _ = col_name_parts
+            new_col_name = col_name.replace(uuid, col_names_map[uuid])
+        return new_col_name
+
     if not streamset._btrdb._ARROW_ENABLED:
         raise NotImplementedError(
             "arrow_to_dataframe requires an arrow-enabled BTrDB server."
@@ -206,35 +219,30 @@ def arrow_to_dataframe(streamset, agg=None, name_callable=None) -> pd.DataFrame:
     if not callable(name_callable):
         name_callable = lambda s: s.collection + "/" + s.name
     # format is: uuid/stat_type
-    tmp_table = streamset.arrow_values()
+    tmp = streamset.arrow_values()
+    # assume time col is the first column
+    time_col = tmp.column_names[0]
+
+    tmp_df = tmp.to_pandas(
+        date_as_object=False,
+        types_mapper=pd.ArrowDtype,
+        split_blocks=True,
+        self_destruct=True,
+    ).set_index(time_col)
+    tmp_df.index.name = time_col
     col_names = _stream_names(streamset, name_callable)
     col_names_map = {str(s.uuid): c for s, c in zip(streamset, col_names)}
-    updated_table_columns = []
-    # assume time col is the first column
-    time_col = tmp_table.column_names[0]
-    for old_col in tmp_table.column_names:
-        if old_col == time_col:
-            updated_table_columns.append(time_col)
-        else:
-            for uu, new_name in col_names_map.items():
-                if uu in old_col:
-                    updated_table_columns.append(old_col.replace(uu, new_name))
-                else:
-                    continue
-    tmp_table = tmp_table.rename_columns(updated_table_columns)
+    tmp_df.rename(
+        columns=lambda col_name: _rename(col_name, col_names_map), inplace=True
+    )
     if not streamset.allow_window:
         usable_cols = []
-        for column_str in tmp_table.column_names:
+        for column_str in tmp_df.columns:
             for agg_name in agg:
                 if agg_name in column_str:
                     usable_cols.append(column_str)
-        tmp = tmp_table.select([time_col, *usable_cols])
-    else:
-        tmp = tmp_table
-    tmp_df = tmp.to_pandas(date_as_object=False, types_mapper=pd.ArrowDtype).set_index(
-        time_col
-    )
-    tmp_df.index.name = time_col
+        tmp_df = tmp_df.loc[:, usable_cols]
+
     return tmp_df