Merge pull request #44 from CSOgroup/sparse-error

Raise ValueError on sparse array

src/cellcharter/tl/_gmm.py

@@ -6,6 +6,7 @@
 import anndata as ad
 import numpy as np
 import pandas as pd
+import scipy.sparse as sps
 import torch
 from lightkit.data import DataLoader, TensorLike, collate_tensor, dataset_from_tensors
 from pycave import set_logging_level
@@ -103,6 +104,11 @@ def fit(self, data: TensorLike) -> GaussianMixture:
         ----------
             The fitted Gaussian mixture.
         """
+        if sps.issparse(data):
+            raise ValueError(
+                "Sparse data is not supported. You may have forgotten to reduce the dimensionality of the data. Otherwise, please convert the data to a dense format."
+            )
+
         if self.init_strategy == "sklearn":
             if self.batch_size is None:
                 kmeans = KMeans(n_clusters=self.num_components, random_state=self.random_state, n_init=1)

tests/tools/test_gmm.py

@@ -0,0 +1,22 @@
+import pytest
+import scipy.sparse as sps
+import squidpy as sq
+
+import cellcharter as cc
+
+
+class TestCluster:
+    @pytest.mark.parametrize("dataset_name", ["mibitof"])
+    def test_sparse(self, dataset_name: str):
+        download_dataset = getattr(sq.datasets, dataset_name)
+        adata = download_dataset()
+        adata.X = sps.csr_matrix(adata.X)
+
+        sq.gr.spatial_neighbors(adata, coord_type="generic", delaunay=True)
+        cc.gr.remove_long_links(adata)
+
+        gmm = cc.tl.Cluster(n_clusters=(10))
+
+        # Check if fit raises a ValueError
+        with pytest.raises(ValueError):
+            gmm.fit(adata, use_rep=None)