Merge pull request #26 from PixelgenTechnologies/feature/exe-1177-fix-filtering-edgelist-bug

Fix filtering edgelist bug, and layout generation problem

Author: johandahlberg

CHANGELOG.md

@@ -5,7 +5,15 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
-## [0.15.0]- 2023-10-16
+## [UNRELEASED]
+
+### Fixed
+
+* Fixed a bug in filtering pixeldataset causing it to return the wrong types.
+* Fixed a bug in graph layout generation due to incorrect data frame concatenation.
+
+
+## [0.15.0] - 2023-10-16
 
 ### Added
 

src/pixelator/graph/backends/implementations.py

@@ -325,6 +325,7 @@ def layout_coordinates(
         coordinates = pd.DataFrame(
             layout_inst.coords,
             columns=["x", "y"] if layout_inst.dim == 2 else ["x", "y", "z"],
+            index=raw.vs["name"],
         )
 
         # If we are doing a 3D layout we add the option of normalized
@@ -340,7 +341,7 @@ def layout_coordinates(
             # Added here to avoid circular imports
             from pixelator.graph.utils import create_node_markers_counts
 
-            node_marker_counts = create_node_markers_counts(raw)
+            node_marker_counts = create_node_markers_counts(self._raw)
             df = pd.concat([coordinates, node_marker_counts], axis=1)
         else:
             df = coordinates

src/pixelator/pixeldataset.py

@@ -700,7 +700,7 @@ def graph(
                 (
                     self.edgelist_lazy.filter(pl.col("component") == component_id)
                     .collect()
-                    .to_pandas(use_pyarrow_extension_array=True)
+                    .to_pandas()
                 )
             )
             if potential_component.empty:
@@ -839,9 +839,7 @@ def _all_true_array(shape):
                 else self.edgelist_lazy
             )
 
-            edgelist = _enforce_edgelist_types(
-                edgelist_pred.collect().to_pandas(use_pyarrow_extension_array=True)
-            )
+            edgelist = _enforce_edgelist_types(edgelist_pred.collect().to_pandas())
 
         if self.polarization is not None:
             polarization_mask = (
@@ -1299,6 +1297,11 @@ def _enforce_edgelist_types(edgelist: pd.DataFrame) -> pd.DataFrame:
     if edgelist.shape[0] == 0:
         edgelist = pd.DataFrame(columns=required_types.keys())
 
+    # If we have the optional sample column, this should be
+    # set to use a categorical type
+    if "sample" in edgelist.columns:
+        required_types["sample"] = "category"
+
     # If all of the prescribed types are already set, just return the edgelist
     type_dict = edgelist.dtypes.to_dict()
     if all(type_dict[key] == type_ for key, type_ in required_types.items()):

tests/graph/test_graph_utils.py

@@ -93,6 +93,76 @@ def test_build_graph_a_node_projected(full_graph_edgelist: pd.DataFrame):
     assert graph.vs.attributes() == ["name", "markers", "type", "pixel_type"]
 
 
+def test_layout_coordinates_all_pixels(full_graph_edgelist: pd.DataFrame):
+    graph = Graph.from_edgelist(
+        edgelist=full_graph_edgelist,
+        add_marker_counts=True,
+        simplify=True,
+        use_full_bipartite=True,
+    )
+    result = graph.layout_coordinates(only_keep_a_pixels=False)
+    assert result.shape == (100, 4)
+    assert set(result.columns) == {"x", "y", "A", "B"}
+
+
+def test_layout_coordinates_3d_layout(full_graph_edgelist: pd.DataFrame):
+    graph = Graph.from_edgelist(
+        edgelist=full_graph_edgelist,
+        add_marker_counts=True,
+        simplify=True,
+        use_full_bipartite=True,
+    )
+    result = graph.layout_coordinates(
+        layout_algorithm="fruchterman_reingold_3d", only_keep_a_pixels=False
+    )
+    assert set(result.columns) == {
+        "x",
+        "y",
+        "z",
+        "x_norm",
+        "y_norm",
+        "z_norm",
+        "A",
+        "B",
+    }
+    assert result.shape == (100, 8)
+
+
+def test_layout_coordinates_only_a_pixels(full_graph_edgelist: pd.DataFrame):
+    graph = Graph.from_edgelist(
+        edgelist=full_graph_edgelist,
+        add_marker_counts=True,
+        simplify=True,
+        use_full_bipartite=True,
+    )
+    result = graph.layout_coordinates(only_keep_a_pixels=True)
+    assert result.shape == (50, 4)
+    assert set(result.columns) == {"x", "y", "A", "B"}
+
+
+def test_layout_coordinates_3d_layout_only_a_pixels(full_graph_edgelist: pd.DataFrame):
+    graph = Graph.from_edgelist(
+        edgelist=full_graph_edgelist,
+        add_marker_counts=True,
+        simplify=True,
+        use_full_bipartite=True,
+    )
+    result = graph.layout_coordinates(
+        layout_algorithm="fruchterman_reingold_3d", only_keep_a_pixels=True
+    )
+    assert set(result.columns) == {
+        "x",
+        "y",
+        "z",
+        "x_norm",
+        "y_norm",
+        "z_norm",
+        "A",
+        "B",
+    }
+    assert result.shape == (50, 8)
+
+
 def test_components_metrics(full_graph_edgelist: pd.DataFrame):
     """Test generating component metrics."""
     # test component metrics

tests/test_pixeldataset.py

@@ -516,7 +516,6 @@ def test_edgelist_to_anndata(
     assert set(adata.obs_names) == set(edgelist["component"].unique())
 
 
-@pytest.mark.test_this
 def test_simple_aggregate(setup_basic_pixel_dataset):
     """test_simple_aggregate."""
     dataset_1, *_ = setup_basic_pixel_dataset
@@ -699,6 +698,23 @@ def test_simple_aggregate_ignore_edgelist(setup_basic_pixel_dataset):
     assert result.edgelist.shape == (0, 9)
 
 
+def test_filter_should_return_proper_typed_edgelist_data(setup_basic_pixel_dataset):
+    # Test to check for bug EXE-1177
+    # This bug was caused by filtering returning an incorrectly typed
+    # edgelist, which in turn caused getting the graph to fail
+    dataset_1, *_ = setup_basic_pixel_dataset
+    dataset_2 = dataset_1.copy()
+
+    aggregated_data = simple_aggregate(
+        sample_names=["sample1", "sample2"], datasets=[dataset_1, dataset_2]
+    )
+
+    result = aggregated_data.filter(components=aggregated_data.adata.obs.index[:2])
+    assert isinstance(result.edgelist["component"].dtype, pd.CategoricalDtype)
+    # Running graph here to make sure it does not raise an exception
+    result.graph(result.adata.obs.index[0])
+
+
 def test_copy(setup_basic_pixel_dataset):
     """test_copy."""
     dataset_1, *_ = setup_basic_pixel_dataset