Merge pull request #2209 from canonizer/bug-fil-bench

[REVIEW] FIL benchmark now works again with gpuarray-c input type

CHANGELOG.md

@@ -71,6 +71,7 @@
 - PR #2494: Set QN regularization strength consistent with scikit-learn
 - PR #2486: Fix cupy input to kmeans init
 - PR #2497: Changes to accomodate cuDF unsigned categorical changes
+- PR #2209: Fix FIL benchmark for gpuarray-c input
 - PR #2507: Import `treelite.sklearn`
 - PR #2521: Fixing invalid smem calculation in KNeighborsCLassifier
 - PR #2515: Increase tolerance for LogisticRegression test

python/cuml/benchmark/bench_helper_funcs.py

@@ -174,15 +174,11 @@ def _build_treelite_classifier(m, data, args, tmpdir):
 
 def _treelite_fil_accuracy_score(y_true, y_pred):
     """Function to get correct accuracy for FIL (returns class index)"""
-    y_pred_binary = input_utils.convert_dtype(y_pred > 0.5, np.int32)
-    if isinstance(y_true, np.ndarray):
-        return cuml.metrics.accuracy_score(y_true, y_pred_binary)
-    elif cuda.devicearray.is_cuda_ndarray(y_true):
-        y_true_np = y_true.copy_to_host()
-        return cuml.metrics.accuracy_score(y_true_np, y_pred_binary)
-    elif isinstance(y_true, cudf.Series):
-        return cuml.metrics.accuracy_score(y_true, y_pred_binary)
-    elif isinstance(y_true, pd.Series):
-        return cuml.metrics.accuracy_score(y_true, y_pred_binary)
-    else:
-        raise TypeError("Received unsupported input type")
+    # convert the input if necessary
+    y_pred1 = (y_pred.copy_to_host() if
+               cuda.devicearray.is_cuda_ndarray(y_pred) else y_pred)
+    y_true1 = (y_true.copy_to_host() if
+               cuda.devicearray.is_cuda_ndarray(y_true) else y_true)
+
+    y_pred_binary = input_utils.convert_dtype(y_pred1 > 0.5, np.int32)
+    return cuml.metrics.accuracy_score(y_true1, y_pred_binary)

python/cuml/test/test_benchmark.py

@@ -190,6 +190,22 @@ def test_real_algos_runner(algo_name):
     assert results["cuml_acc"] is not None
 
 
+# Test FIL with several input types
+@pytest.mark.parametrize('input_type', ['numpy', 'cudf', 'gpuarray',
+                                        'gpuarray-c'])
+def test_fil_input_types(input_type):
+    pair = algorithms.algorithm_by_name('FIL')
+
+    if not has_xgboost():
+        pytest.xfail()
+
+    runner = AccuracyComparisonRunner(
+        [20], [5], dataset_name='classification', test_fraction=0.5,
+        input_type=input_type)
+    results = runner.run(pair, run_cpu=False)[0]
+    assert results["cuml_acc"] is not None
+
+
 @pytest.mark.parametrize('input_type', ['numpy', 'cudf', 'pandas', 'gpuarray'])
 def test_training_data_to_numpy(input_type):
     X, y, *_ = datagen.gen_data(