(feat): use anndata_to_GPU

Author: ilan-gold

benchmarks/benchmarks/preprocessing.py

@@ -16,7 +16,7 @@ class PreprocessingSuite:
     param_names = ["input_data"]
 
     def setup(self, input_data: str):
-        self.adata = self._data_dict[input_data].copy()
+        self.adata = rsc.get.anndata_to_GPU(self._data_dict[input_data].copy(), copy=True)
 
     def time_calculate_qc_metrics(self, *_):
         self.adata.var["mt"] = self.adata.var_names.str.startswith("MT-")

benchmarks/benchmarks/readwrite.py

@@ -34,26 +34,25 @@
     from collections.abc import Callable
     from pathlib import Path
 
+import pathlib
+
+sc.settings.datasetdir = pathlib.Path(__file__).parent.resolve() / "data"
 
 @dataclass
 class Dataset:
     path: Path
     get: Callable[[], anndata.AnnData]
 
-
-pbmc3k = Dataset(
-    path=sc.settings.datasetdir / "pbmc3k_raw.h5ad", get=sc.datasets.pbmc3k
-)
+path="/p/project/training2406/team_scverse/gold2/rapids_singlecell/benchmarks/data/pbmc3k_raw.h5ad"
 
 
 class ToGPUSuite:
-    _data_dict = dict(pbmc3k=pbmc3k)
+    _data_dict = dict(pbmc3k=anndata.read_h5ad(path))
     params = _data_dict.keys()
     param_names = ["input_data"]
 
     def setup(self, input_data: str):
-        self.path = self._data_dict[input_data].path
-        self.data = self._data_dict[input_data].get()
+        self.data = self._data_dict[input_data]
 
     def time_to_gpu(self, *_):
         anndata_to_GPU(self.data)

benchmarks/benchmarks/squidpy.py

@@ -7,21 +7,21 @@
 
 from itertools import product
 
-import scanpy as sc
+import anndata as ad
 
 import rapids_singlecell as rsc
 
+import pathlib
 
 class ToolsSuite:
     _data_dict = dict(
-        visium_sge=sc.datasets.visium_sge(),
+        visium_sge=ad.read_h5ad("/p/project/training2406/team_scverse/gold2/rapids_singlecell/benchmarks/data/paul15.h5ad"),
     )
     params = _data_dict.keys()
     param_names = ["input_data"]
 
     def setup(self, input_data):
-        self.adata = self._data_dict[input_data].copy()
-        assert "X_pca" in self.adata.obsm
+        self.adata = rsc.get.anndata_to_GPU(self._data_dict[input_data].copy(), copy=True)
 
     def time_ligrec(self, *_):
         gene_ids = self.adata.var.index
@@ -31,8 +31,7 @@ def time_ligrec(self, *_):
             "leiden",
             interactions=interactions,
             n_perms=5,
-            use_raw=True,
-            copy=True,
+            use_raw=False,
         )
 
     def peakmem_ligrec(self, *_):
@@ -43,8 +42,7 @@ def peakmem_ligrec(self, *_):
             "leiden",
             interactions=interactions,
             n_perms=5,
-            use_raw=True,
-            copy=True,
+            use_raw=False,
         )
 
     def time_autocorr_moran(self, *_):

benchmarks/benchmarks/tools.py

@@ -9,6 +9,8 @@
 
 import rapids_singlecell as rsc
 
+import pathlib
+
 
 class ToolsSuite:
     _data_dict = dict(
@@ -18,7 +20,7 @@ class ToolsSuite:
     param_names = ["input_data"]
 
     def setup(self, input_data):
-        self.adata = self._data_dict[input_data].copy()
+        self.adata = rsc.get.anndata_to_GPU(self._data_dict[input_data].copy(), copy=True)
         assert "X_pca" in self.adata.obsm
 
     def time_umap(self, *_):
@@ -40,7 +42,7 @@ def peakmem_leiden(self, *_):
         rsc.tl.leiden(self.adata)
 
     def time_embedding_denity(self, *_):
-        rsc.tl.embedding_density(self.adata, basis="X_umap")
+        rsc.tl.embedding_density(self.adata, basis="umap")
 
     def peakmem_embedding_denity(self, *_):
-        rsc.tl.embedding_density(self.adata, basis="X_umap")
+        rsc.tl.embedding_density(self.adata, basis="umap")

benchmarks/environment.yml

@@ -1,11 +1,10 @@
-name: test_asv
 channels:
   - rapidsai
   - nvidia
   - conda-forge
 dependencies:
   - python=3.11
-  - cuda-version=11.8
+  - cuda-version=12
   - cudf=24.4
   - cuml=24.4
   - cugraph=24.4