core-lens 0.1.dev148__tar.gz → 0.1.dev150__tar.gz

{core_lens-0.1.dev148 → core_lens-0.1.dev150}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: core-lens
-Version: 0.1.dev148
+Version: 0.1.dev150
 Summary: Query, analyse, and visualise CoreStack's microwatershed and Earth science data through a clean, composable Python API.
 Project-URL: Homepage, https://github.com/ApoorvaKashyap/core-lens
 Project-URL: Issues, https://github.com/ApoorvaKashyap/core-lens/issues

{core_lens-0.1.dev148 → core_lens-0.1.dev150}/src/core_lens/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '0.1.dev148'
-__version_tuple__ = version_tuple = (0, 1, 'dev148')
+__version__ = version = '0.1.dev150'
+__version_tuple__ = version_tuple = (0, 1, 'dev150')
 
 __commit_id__ = commit_id = None

{core_lens-0.1.dev148 → core_lens-0.1.dev150}/src/core_lens/base/result.py

@@ -155,10 +155,13 @@ class Result:
             return self
 
         key_cols = self.key_cols
-        geo_lf = self.entity.geometry_lazy.join(
-            self.data.select(key_cols), on=key_cols, how="semi"
-        )
-        joined = self.data.join(geo_lf, on=key_cols, how="left")
+
+        # We use a direct left join here instead of a semi-join intermediate step.
+        # A semi-join creates a "diamond" query graph where `self.data` must be evaluated
+        # twice by Polars. Because `self.data` can be an expensive aggregation pipeline,
+        # this caused execution times to double.
+        joined = self.data.join(self.entity.geometry_lazy, on=key_cols, how="left")
+
         return self._replace(data=joined, has_geometry=True)
 
     def derive(self, name: str, expr: pl.Expr) -> "Result":

{core_lens-0.1.dev148 → core_lens-0.1.dev150}/src/core_lens/utils/polars_utils.py

@@ -140,7 +140,15 @@ def scan_with_key_filter(
     if time_expr is not None:
         lf = lf.filter(time_expr)
 
-    kv_lazy = key_values.lazy() if isinstance(key_values, pl.DataFrame) else key_values
-    lf = lf.join(kv_lazy, on=key_cols, how="semi")
+    kv_df = key_values.collect() if isinstance(key_values, pl.LazyFrame) else key_values
+    if len(key_cols) == 1 and kv_df.height < 50000:
+        # Fast path: For reasonably sized AoIs, `is_in` enables perfect row-group
+        # predicate pushdown in the Parquet reader, avoiding a full table scan.
+        # The literal tree RAM issue only occurs with hundreds of thousands of keys.
+        key = key_cols[0]
+        lf = lf.filter(pl.col(key).is_in(kv_df[key]))
+    else:
+        # Fallback to semi-join for massive key arrays (e.g. state-wide scopes)
+        lf = lf.join(kv_df.lazy(), on=key_cols, how="semi")
 
     return lf