Enable pytest failures on warnings from cudf (#4223)

In 24.04, cudf issues a FutureWarning when using positional indexing with `__getitem__`. The change here is to use `iloc` instead.

Additionally the test suite will now fail when a warning comes from cudf as apart of rapidsai/build-planning#26

Authors:
  - Matthew Roeschke (https://github.com/mroeschke)
  - Ralph Liu (https://github.com/nv-rliu)
  - Brad Rees (https://github.com/BradReesWork)

Approvers:
  - Rick Ratzel (https://github.com/rlratzel)
  - Don Acosta (https://github.com/acostadon)

URL: #4223

python/cugraph/cugraph/centrality/betweenness_centrality.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2019-2023, NVIDIA CORPORATION.
+# Copyright (c) 2019-2024, NVIDIA CORPORATION.
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
@@ -160,7 +160,7 @@ def betweenness_centrality(
 
     if not isinstance(k, (cudf.DataFrame, cudf.Series)):
         if isinstance(k, list):
-            vertex_dtype = G.edgelist.edgelist_df.dtypes[0]
+            vertex_dtype = G.edgelist.edgelist_df.dtypes.iloc[0]
             k = cudf.Series(k, dtype=vertex_dtype)
 
     if isinstance(k, (cudf.DataFrame, cudf.Series)):
@@ -300,7 +300,7 @@ def edge_betweenness_centrality(
 
     if not isinstance(k, (cudf.DataFrame, cudf.Series)):
         if isinstance(k, list):
-            vertex_dtype = G.edgelist.edgelist_df.dtypes[0]
+            vertex_dtype = G.edgelist.edgelist_df.dtypes.iloc[0]
             k = cudf.Series(k, dtype=vertex_dtype)
 
     if isinstance(k, (cudf.DataFrame, cudf.Series)):

python/cugraph/cugraph/community/spectral_clustering.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2019-2023, NVIDIA CORPORATION.
+# Copyright (c) 2019-2024, NVIDIA CORPORATION.
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
@@ -92,8 +92,8 @@ def spectralBalancedCutClustering(
     G, isNx = ensure_cugraph_obj_for_nx(G)
     # Check if vertex type is "int32"
     if (
-        G.edgelist.edgelist_df.dtypes[0] != np.int32
-        or G.edgelist.edgelist_df.dtypes[1] != np.int32
+        G.edgelist.edgelist_df.dtypes.iloc[0] != np.int32
+        or G.edgelist.edgelist_df.dtypes.iloc[1] != np.int32
     ):
         raise ValueError(
             "'spectralBalancedCutClustering' requires the input graph's vertex to be "
@@ -186,8 +186,8 @@ def spectralModularityMaximizationClustering(
 
     G, isNx = ensure_cugraph_obj_for_nx(G)
     if (
-        G.edgelist.edgelist_df.dtypes[0] != np.int32
-        or G.edgelist.edgelist_df.dtypes[1] != np.int32
+        G.edgelist.edgelist_df.dtypes.iloc[0] != np.int32
+        or G.edgelist.edgelist_df.dtypes.iloc[1] != np.int32
     ):
         raise ValueError(
             "'spectralModularityMaximizationClustering' requires the input graph's "
@@ -271,8 +271,8 @@ def analyzeClustering_modularity(
 
     G, isNx = ensure_cugraph_obj_for_nx(G)
     if (
-        G.edgelist.edgelist_df.dtypes[0] != np.int32
-        or G.edgelist.edgelist_df.dtypes[1] != np.int32
+        G.edgelist.edgelist_df.dtypes.iloc[0] != np.int32
+        or G.edgelist.edgelist_df.dtypes.iloc[1] != np.int32
     ):
         raise ValueError(
             "'analyzeClustering_modularity' requires the input graph's "
@@ -284,7 +284,7 @@ def analyzeClustering_modularity(
             clustering, "vertex", vertex_col_name, drop=True
         )
 
-    if clustering.dtypes[0] != np.int32 or clustering.dtypes[1] != np.int32:
+    if clustering.dtypes.iloc[0] != np.int32 or clustering.dtypes.iloc[1] != np.int32:
         raise ValueError(
             "'analyzeClustering_modularity' requires both the clustering 'vertex' "
             "and 'cluster' to be of type 'int32'"
@@ -354,8 +354,8 @@ def analyzeClustering_edge_cut(
     G, isNx = ensure_cugraph_obj_for_nx(G)
 
     if (
-        G.edgelist.edgelist_df.dtypes[0] != np.int32
-        or G.edgelist.edgelist_df.dtypes[1] != np.int32
+        G.edgelist.edgelist_df.dtypes.iloc[0] != np.int32
+        or G.edgelist.edgelist_df.dtypes.iloc[1] != np.int32
     ):
         raise ValueError(
             "'analyzeClustering_edge_cut' requires the input graph's vertex to be "
@@ -367,7 +367,7 @@ def analyzeClustering_edge_cut(
             clustering, "vertex", vertex_col_name, drop=True
         )
 
-    if clustering.dtypes[0] != np.int32 or clustering.dtypes[1] != np.int32:
+    if clustering.dtypes.iloc[0] != np.int32 or clustering.dtypes.iloc[1] != np.int32:
         raise ValueError(
             "'analyzeClustering_edge_cut' requires both the clustering 'vertex' "
             "and 'cluster' to be of type 'int32'"
@@ -437,7 +437,7 @@ def analyzeClustering_ratio_cut(
             clustering, "vertex", vertex_col_name, drop=True
         )
 
-    if clustering.dtypes[0] != np.int32 or clustering.dtypes[1] != np.int32:
+    if clustering.dtypes.iloc[0] != np.int32 or clustering.dtypes.iloc[1] != np.int32:
         raise ValueError(
             "'analyzeClustering_ratio_cut' requires both the clustering 'vertex' "
             "and 'cluster' to be of type 'int32'"

python/cugraph/cugraph/community/triangle_count.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2019-2023, NVIDIA CORPORATION.
+# Copyright (c) 2019-2024, NVIDIA CORPORATION.
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
@@ -24,7 +24,7 @@
 # FIXME: Move this function to the utility module so that it can be
 # shared by other algos
 def ensure_valid_dtype(input_graph, start_list):
-    vertex_dtype = input_graph.edgelist.edgelist_df.dtypes[0]
+    vertex_dtype = input_graph.edgelist.edgelist_df.dtypes.iloc[0]
     if isinstance(start_list, cudf.Series):
         start_list_dtypes = start_list.dtype
     else:

python/cugraph/cugraph/dask/community/leiden.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2022-2023, NVIDIA CORPORATION.
+# Copyright (c) 2022-2024, NVIDIA CORPORATION.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -170,7 +170,7 @@ def leiden(
     part_mod_score = [client.submit(convert_to_cudf, r) for r in result]
     wait(part_mod_score)
 
-    vertex_dtype = input_graph.edgelist.edgelist_df.dtypes[0]
+    vertex_dtype = input_graph.edgelist.edgelist_df.dtypes.iloc[0]
     empty_df = cudf.DataFrame(
         {
             "vertex": numpy.empty(shape=0, dtype=vertex_dtype),

python/cugraph/cugraph/dask/community/louvain.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2022-2023, NVIDIA CORPORATION.
+# Copyright (c) 2022-2024, NVIDIA CORPORATION.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -191,7 +191,7 @@ def louvain(
     part_mod_score = [client.submit(convert_to_cudf, r) for r in result]
     wait(part_mod_score)
 
-    vertex_dtype = input_graph.edgelist.edgelist_df.dtypes[0]
+    vertex_dtype = input_graph.edgelist.edgelist_df.dtypes.iloc[0]
     empty_df = cudf.DataFrame(
         {
             "vertex": numpy.empty(shape=0, dtype=vertex_dtype),

python/cugraph/cugraph/dask/link_analysis/pagerank.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2019-2023, NVIDIA CORPORATION.
+# Copyright (c) 2019-2024, NVIDIA CORPORATION.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -76,7 +76,7 @@ def ensure_valid_dtype(input_graph, input_df, input_df_name):
             warnings.warn(warning_msg, UserWarning)
             input_df = input_df.astype({"values": edge_attr_dtype})
 
-    vertex_dtype = input_graph.edgelist.edgelist_df.dtypes[0]
+    vertex_dtype = input_graph.edgelist.edgelist_df.dtypes.iloc[0]
     input_df_vertex_dtype = input_df["vertex"].dtype
     if input_df_vertex_dtype != vertex_dtype:
         warning_msg = (
@@ -406,7 +406,7 @@ def pagerank(
 
     wait(result)
 
-    vertex_dtype = input_graph.edgelist.edgelist_df.dtypes[0]
+    vertex_dtype = input_graph.edgelist.edgelist_df.dtypes.iloc[0]
 
     # Have each worker convert tuple of arrays and bool from PLC to cudf
     # DataFrames and bools. This will be a list of futures.

python/cugraph/cugraph/dask/sampling/random_walks.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2022-2023, NVIDIA CORPORATION.
+# Copyright (c) 2022-2024, NVIDIA CORPORATION.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -123,7 +123,7 @@ def random_walks(
         start_vertices_type = input_graph.edgelist.edgelist_df.dtypes[0]
     else:
         # FIXME: Get the 'src' column names instead and retrieve the type
-        start_vertices_type = input_graph.input_df.dtypes[0]
+        start_vertices_type = input_graph.input_df.dtypes.iloc[0]
     start_vertices = dask_cudf.from_cudf(
         start_vertices, npartitions=min(input_graph._npartitions, len(start_vertices))
     )

python/cugraph/cugraph/gnn/dgl_extensions/dgl_uniform_sampler.py

@@ -150,7 +150,7 @@ def _get_type_id_from_indices(indices, etype_id_range_dict):
 
         for etype_id, (start, stop) in etype_id_range_dict.items():
             range_types = (start <= indices) & (indices < stop)
-            type_ser[range_types] = etype_id
+            type_ser[range_types] = type_ser.dtype.type(etype_id)
 
         return type_ser
 

python/cugraph/cugraph/link_analysis/pagerank.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2019-2023, NVIDIA CORPORATION.
+# Copyright (c) 2019-2024, NVIDIA CORPORATION.
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
@@ -64,7 +64,7 @@ def ensure_valid_dtype(input_graph, input_df, input_df_name):
             warnings.warn(warning_msg, UserWarning)
             input_df = input_df.astype({"values": edge_attr_dtype})
 
-    vertex_dtype = input_graph.edgelist.edgelist_df.dtypes[0]
+    vertex_dtype = input_graph.edgelist.edgelist_df.dtypes.iloc[0]
     input_df_vertex_dtype = input_df["vertex"].dtype
     if input_df_vertex_dtype != vertex_dtype:
         warning_msg = (

python/cugraph/cugraph/link_prediction/jaccard.py

@@ -38,10 +38,13 @@
 # FIXME: Move this function to the utility module so that it can be
 # shared by other algos
 def ensure_valid_dtype(input_graph, vertex_pair):
-    vertex_dtype = input_graph.edgelist.edgelist_df.dtypes[0]
+    vertex_dtype = input_graph.edgelist.edgelist_df.dtypes.iloc[0]
     vertex_pair_dtypes = vertex_pair.dtypes
 
-    if vertex_pair_dtypes[0] != vertex_dtype or vertex_pair_dtypes[1] != vertex_dtype:
+    if (
+        vertex_pair_dtypes.iloc[0] != vertex_dtype
+        or vertex_pair_dtypes.iloc[1] != vertex_dtype
+    ):
         warning_msg = (
             "Jaccard requires 'vertex_pair' to match the graph's 'vertex' type. "
             f"input graph's vertex type is: {vertex_dtype} and got "

python/cugraph/cugraph/link_prediction/overlap.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2019-2023, NVIDIA CORPORATION.
+# Copyright (c) 2019-2024, NVIDIA CORPORATION.
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
@@ -38,10 +38,13 @@
 # FIXME: Move this function to the utility module so that it can be
 # shared by other algos
 def ensure_valid_dtype(input_graph, vertex_pair):
-    vertex_dtype = input_graph.edgelist.edgelist_df.dtypes[0]
+    vertex_dtype = input_graph.edgelist.edgelist_df.dtypes.iloc[0]
     vertex_pair_dtypes = vertex_pair.dtypes
 
-    if vertex_pair_dtypes[0] != vertex_dtype or vertex_pair_dtypes[1] != vertex_dtype:
+    if (
+        vertex_pair_dtypes.iloc[0] != vertex_dtype
+        or vertex_pair_dtypes.iloc[1] != vertex_dtype
+    ):
         warning_msg = (
             "Overlap requires 'vertex_pair' to match the graph's 'vertex' type. "
             f"input graph's vertex type is: {vertex_dtype} and got "

python/cugraph/cugraph/link_prediction/sorensen.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2021-2023, NVIDIA CORPORATION.
+# Copyright (c) 2021-2024, NVIDIA CORPORATION.
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
@@ -38,10 +38,13 @@
 # FIXME: Move this function to the utility module so that it can be
 # shared by other algos
 def ensure_valid_dtype(input_graph, vertex_pair):
-    vertex_dtype = input_graph.edgelist.edgelist_df.dtypes[0]
+    vertex_dtype = input_graph.edgelist.edgelist_df.dtypes.iloc[0]
     vertex_pair_dtypes = vertex_pair.dtypes
 
-    if vertex_pair_dtypes[0] != vertex_dtype or vertex_pair_dtypes[1] != vertex_dtype:
+    if (
+        vertex_pair_dtypes.iloc[0] != vertex_dtype
+        or vertex_pair_dtypes.iloc[1] != vertex_dtype
+    ):
         warning_msg = (
             "Sorensen requires 'vertex_pair' to match the graph's 'vertex' type. "
             f"input graph's vertex type is: {vertex_dtype} and got "

python/cugraph/cugraph/sampling/random_walks.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2022-2023, NVIDIA CORPORATION.
+# Copyright (c) 2022-2024, NVIDIA CORPORATION.
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
@@ -145,7 +145,7 @@ def random_walks(
     if isinstance(start_vertices, list):
         # Ensure the 'start_vertices' have the same dtype as the edge list.
         # Failing to do that may produce erroneous results.
-        vertex_dtype = G.edgelist.edgelist_df.dtypes[0]
+        vertex_dtype = G.edgelist.edgelist_df.dtypes.iloc[0]
         start_vertices = cudf.Series(start_vertices, dtype=vertex_dtype)
 
     if G.renumbered is True:

python/cugraph/cugraph/sampling/uniform_neighbor_sample.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2022-2023, NVIDIA CORPORATION.
+# Copyright (c) 2022-2024, NVIDIA CORPORATION.
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
@@ -38,11 +38,11 @@
 # FIXME: Move this function to the utility module so that it can be
 # shared by other algos
 def ensure_valid_dtype(input_graph, start_list):
-    vertex_dtype = input_graph.edgelist.edgelist_df.dtypes[0]
+    vertex_dtype = input_graph.edgelist.edgelist_df.dtypes.iloc[0]
     if isinstance(start_list, cudf.Series):
         start_list_dtypes = start_list.dtype
     else:
-        start_list_dtypes = start_list.dtypes[0]
+        start_list_dtypes = start_list.dtypes.iloc[0]
 
     if start_list_dtypes != vertex_dtype:
         warning_msg = (

python/cugraph/cugraph/tests/centrality/test_edge_betweenness_centrality_mg.py

@@ -53,6 +53,7 @@ def get_sg_graph(dataset, directed, edge_ids):
         if not directed:
             # Edge ids not supported for undirected graph
             return None
+
         dtype = df.dtypes.iloc[0]
         edge_id = "edge_id"
         df[edge_id] = df.index
@@ -79,7 +80,7 @@ def get_mg_graph(dataset, directed, edge_ids, weight):
         weight = None
 
     if edge_ids:
-        dtype = ddf.dtypes[0]
+        dtype = ddf.dtypes.iloc[0]
         edge_id = "edge_id"
         ddf = ddf.assign(idx=1)
         ddf["edge_id"] = ddf.idx.cumsum().astype(dtype) - 1

python/cugraph/cugraph/tests/layout/test_force_atlas2.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2020-2023, NVIDIA CORPORATION.
+# Copyright (c) 2020-2024, NVIDIA CORPORATION.
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
@@ -122,9 +122,9 @@ def cugraph_call(
 BARNES_HUT_OPTIMIZE = [False, True]
 
 
-class TestCallback(GraphBasedDimRedCallback):
+class ExampleCallback(GraphBasedDimRedCallback):
     def __init__(self):
-        super(TestCallback, self).__init__()
+        super().__init__()
         self.on_preprocess_end_called_count = 0
         self.on_epoch_end_called_count = 0
         self.on_train_end_called_count = 0
@@ -145,7 +145,7 @@ def on_train_end(self, positions):
 @pytest.mark.parametrize("barnes_hut_optimize", BARNES_HUT_OPTIMIZE)
 def test_force_atlas2(graph_file, score, max_iter, barnes_hut_optimize):
     cu_M = graph_file.get_edgelist(download=True)
-    test_callback = TestCallback()
+    test_callback = ExampleCallback()
     cu_pos = cugraph_call(
         cu_M,
         max_iter=max_iter,

python/cugraph/pytest.ini

@@ -1,4 +1,4 @@
-# Copyright (c) 2021-2022, NVIDIA CORPORATION.
+# Copyright (c) 2021-2024, NVIDIA CORPORATION.
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
@@ -56,3 +56,8 @@ python_files =
 python_functions =
           bench_*
           test_*
+
+filterwarnings =
+          error:::cudf
+          # Called via dask. Not obviously addressable in cugraph.
+          ignore:The behavior of array concatenation with empty entries is deprecated:FutureWarning:cudf