Further fixes for Scipy 1.15 update for PR and nightly CI (#6213)

Nightly CI revealed a bug between hypothesis `floating_dtypes(sizes=(32, 64)` and building sparse matrices, this PR uses `st.sampled_from((np.float32, np.float64)` to solve the issue. 

Additionally, cudf.pandas active made one dataset in ARIMA pytests fail, so disabling that one while we look further into it.

Authors:
  - Dante Gama Dessavre (https://github.com/dantegd)

Approvers:
  - Victor Lafargue (https://github.com/viclafargue)
  - Simon Adorf (https://github.com/csadorf)

URL: #6213

python/cuml/cuml/tests/test_arima.py

@@ -36,7 +36,10 @@
 from cuml.internals.safe_imports import gpu_only_import
 import statsmodels.api as sm
 from sklearn.model_selection import train_test_split
-from cuml.internals.safe_imports import cpu_only_import_from
+from cuml.internals.safe_imports import (
+    cpu_only_import_from,
+    gpu_only_import_from,
+)
 import warnings
 import os
 import pytest
@@ -49,6 +52,7 @@
 approx_fprime = cpu_only_import_from("scipy.optimize", "approx_fprime")
 
 cudf = gpu_only_import("cudf")
+cudf_pandas_active = gpu_only_import_from("cudf.pandas", "LOADED")
 
 
 ###############################################################################
@@ -410,6 +414,11 @@ def fill_interpolation(df_in):
 @pytest.mark.parametrize("dtype", [np.float64])
 def test_integration(key, data, dtype):
     """Full integration test: estimate, fit, forecast"""
+    if (
+        data.dataset == "endog_hourly_earnings_by_industry_missing_exog"
+        and cudf_pandas_active
+    ):
+        pytest.skip(reason="https://github.com/rapidsai/cuml/issues/6209")
     order, seasonal_order, intercept = extract_order(key)
     s = max(1, seasonal_order[3])
 

python/cuml/cuml/tests/test_linear_model.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2019-2024, NVIDIA CORPORATION.
+# Copyright (c) 2019-2025, NVIDIA CORPORATION.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -629,7 +629,7 @@ def test_logistic_regression_model_default(dtype):
 
 
 @given(
-    dtype=floating_dtypes(sizes=(32, 64)),
+    dtype=st.sampled_from((np.float32, np.float64)),
     order=st.sampled_from(("C", "F")),
     sparse_input=st.booleans(),
     fit_intercept=st.booleans(),
@@ -661,7 +661,7 @@ def test_logistic_regression_model_digits(
     assert score >= acceptable_score
 
 
-@given(dtype=floating_dtypes(sizes=(32, 64)))
+@given(dtype=st.sampled_from((np.float32, np.float64)))
 def test_logistic_regression_sparse_only(dtype, nlp_20news):
 
     # sklearn score with max_iter = 10000
@@ -685,7 +685,7 @@ def test_logistic_regression_sparse_only(dtype, nlp_20news):
 @given(
     dataset=split_datasets(
         standard_classification_datasets(
-            dtypes=floating_dtypes(sizes=(32, 64)),
+            dtypes=st.sampled_from((np.float32, np.float64)),
             n_classes=st.sampled_from((2, 10)),
             n_features=st.just(20),
             n_informative=st.just(10),
@@ -727,7 +727,7 @@ def test_logistic_regression_decision_function(
 @given(
     dataset=split_datasets(
         standard_classification_datasets(
-            dtypes=floating_dtypes(sizes=(32, 64)),
+            dtypes=st.sampled_from((np.float32, np.float64)),
             n_classes=st.sampled_from((2, 10)),
             n_features=st.just(20),
             n_informative=st.just(10),