diff --git a/quackosm/pbf_file_reader.py b/quackosm/pbf_file_reader.py
index 9ff75ca..4030415 100644
--- a/quackosm/pbf_file_reader.py
+++ b/quackosm/pbf_file_reader.py
@@ -475,15 +475,6 @@ def _parse_pbf_file(
             )
             self._delete_directories("nodes_filtered_ids")
 
-            # ways_refs_with_nodes_structs = self._get_ways_refs_with_nodes_structs(
-            #     converted_osm_parquet_files
-            # )
-            # self._delete_directories(
-            #     [
-            #         "nodes_valid_with_tags",
-            #     ],
-            # )
-
             filtered_ways_with_linestrings = self._get_filtered_ways_with_linestrings(
                 osm_parquet_files=converted_osm_parquet_files
             )
@@ -496,7 +487,6 @@ def _parse_pbf_file(
                     "ways_required_grouped",
                     "ways_required_ids",
                     "ways_with_unnested_nodes_refs",
-                    # "ways_refs_with_nodes_structs",
                     "required_ways_ids_grouped",
                     "required_ways_grouped",
                     "required_ways_tmp",
@@ -775,19 +765,17 @@ def _prefilter_elements_ids(
             ways_valid_ids = self._sql_to_parquet_file(
                 sql_query=f"""
                 WITH total_ways_with_nodes_refs AS (
-                    SELECT id, ref
+                    SELECT id
                     FROM ({ways_with_unnested_nodes_refs.sql_query()})
                 ),
                 unmatched_ways_with_nodes_refs AS (
-                    SELECT id, ref
+                    SELECT id
                     FROM ({ways_with_unnested_nodes_refs.sql_query()}) w
                     ANTI JOIN ({nodes_valid_with_tags.sql_query()}) nv ON nv.id = w.ref
                 )
                 SELECT DISTINCT id
                 FROM total_ways_with_nodes_refs
-                EXCEPT
-                SELECT DISTINCT id
-                FROM unmatched_ways_with_nodes_refs
+                ANTI JOIN unmatched_ways_with_nodes_refs USING (id)
                 """,
                 file_path=self.tmp_dir_path / "ways_valid_ids",
             )
@@ -879,19 +867,17 @@ def _prefilter_elements_ids(
             relations_valid_ids = self._sql_to_parquet_file(
                 sql_query=f"""
                 WITH total_relation_refs AS (
-                    SELECT id, ref
+                    SELECT id
                     FROM ({relations_with_unnested_way_refs.sql_query()}) frr
                 ),
                 unmatched_relation_refs AS (
-                    SELECT id, ref
+                    SELECT id
                     FROM ({relations_with_unnested_way_refs.sql_query()}) r
                     ANTI JOIN ({ways_valid_ids.sql_query()}) wv ON wv.id = r.ref
                 )
                 SELECT DISTINCT id
                 FROM total_relation_refs
-                EXCEPT
-                SELECT DISTINCT id
-                FROM unmatched_relation_refs
+                ANTI JOIN unmatched_relation_refs USING (id)
                 """,
                 file_path=self.tmp_dir_path / "relations_valid_ids",
             )
@@ -1076,7 +1062,7 @@ def _save_parquet_file(
     ) -> "duckdb.DuckDBPyRelation":
         query = f"""
             COPY (
-                SELECT * FROM ({relation.sql_query()})
+                {relation.sql_query()}
             ) TO '{file_path}' (
                 FORMAT 'parquet',
                 PER_THREAD_OUTPUT true,
@@ -1084,9 +1070,13 @@ def _save_parquet_file(
                 COMPRESSION '{self.parquet_compression}'
             )
         """
+        self._run_query(query, run_in_separate_process)
+        return self.connection.sql(f"SELECT * FROM read_parquet('{file_path}/**')")
+
+    def _run_query(self, sql_query: str, run_in_separate_process: bool = False) -> None:
         if run_in_separate_process:
             with Pool() as pool:
-                r = pool.apply_async(_run_query, args=(query, self.tmp_dir_path))
+                r = pool.apply_async(_run_query, args=(sql_query, self.tmp_dir_path))
                 actual_memory = psutil.virtual_memory()
                 percentage_threshold = 95
                 if (actual_memory.total * 0.05) > MEMORY_1GB:
@@ -1102,8 +1092,7 @@ def _save_parquet_file(
                         sleep(0.5)
                 r.get()
         else:
-            self.connection.sql(query)
-        return self.connection.sql(f"SELECT * FROM read_parquet('{file_path}/**')")
+            self.connection.sql(sql_query)
 
     def _calculate_unique_ids_to_parquet(
         self, file_path: Path, result_path: Optional[Path] = None
@@ -1447,6 +1436,8 @@ def _group_ways(
                     """
                 )
 
+                self._delete_directories(current_ways_group_path)
+
         return groups
 
     def _construct_ways_linestrings(
@@ -1458,24 +1449,32 @@ def _construct_ways_linestrings(
     ) -> None:
         for group in bar.track(range(groups + 1)):
             current_ways_group_path = grouped_ways_path / f"group={group}"
-            ways_with_linestrings = self.connection.sql(
-                f"""
-                SELECT id, list(point ORDER BY ref_idx ASC)::LINESTRING_2D linestring
-                FROM read_parquet(
-                    '{current_ways_group_path}/**',
-                    hive_partitioning = true,
-                    hive_types = {{ 'ref_idx': SMALLINT }}
+            current_destination_path = destination_dir_path / f"group={group}"
+
+            query = f"""
+                COPY (
+                    SELECT id, list(point ORDER BY ref_idx ASC)::LINESTRING_2D linestring
+                    FROM read_parquet(
+                        '{current_ways_group_path}/**',
+                        hive_partitioning = TRUE,
+                        hive_types = {{ 'ref_idx': SMALLINT }}
+                    )
+                    GROUP BY id
+                ) TO '{current_destination_path}' (
+                    FORMAT 'parquet',
+                    PER_THREAD_OUTPUT true,
+                    ROW_GROUP_SIZE 25000,
+                    COMPRESSION '{self.parquet_compression}'
                 )
-                GROUP BY id
-                """
-            )
-            self._save_parquet_file(
-                relation=ways_with_linestrings,
-                file_path=destination_dir_path / f"group={group}",
+            """
+
+            self._run_query(
+                query,
                 run_in_separate_process=(
                     self.rows_per_bucket > PbfFileReader.ROWS_PER_BUCKET_MEMORY_CONFIG[0]
                 ),
             )
+
             self._delete_directories(current_ways_group_path)
 
     def _get_filtered_ways_with_proper_geometry(