[FEA] Add GPUTreeSHAP to cuML explainer module (experimental)

cpp/CMakeLists.txt

@@ -201,6 +201,7 @@ if(LINK_FAISS)
 endif()
 
 include(cmake/thirdparty/get_treelite.cmake)
+include(cmake/thirdparty/get_gputreeshap.cmake)
 include(cmake/thirdparty/get_raft.cmake)
 
 if(NOT SINGLEGPU)
@@ -264,6 +265,7 @@ if(BUILD_CUML_CPP_LIBRARY)
         src/decisiontree/decisiontree.cu
         src/explainer/kernel_shap.cu
         src/explainer/permutation_shap.cu
+        src/explainer/tree_shap.cu
         src/fil/fil.cu
         src/fil/infer.cu
         src/glm/glm.cu
@@ -388,6 +390,7 @@ if(BUILD_CUML_CPP_LIBRARY)
       CUDA::cusolver
       CUDA::cudart
       CUDA::cusparse
+      GPUTreeShap::GPUTreeShap
       $<$<BOOL:${NVTX}>:CUDA::nvToolsExt>
       $<$<BOOL:${LINK_FAISS}>:FAISS::FAISS>
       $<IF:$<BOOL:${Treelite_ADDED}>,treelite::treelite_static,treelite::treelite>

cpp/cmake/thirdparty/get_gputreeshap.cmake

@@ -0,0 +1,34 @@
+#=============================================================================
+# Copyright (c) 2021, 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
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#=============================================================================
+
+function(find_and_configure_gputreeshap)
+
+    set(oneValueArgs VERSION PINNED_TAG)
+    cmake_parse_arguments(PKG "${options}" "${oneValueArgs}"
+                          "${multiValueArgs}" ${ARGN} )
+
+    rapids_cpm_find(GPUTreeShap 0.0.1
+        GLOBAL_TARGETS  GPUTreeShap::GPUTreeShap GPUTreeShap
+        CPM_ARGS
+            GIT_REPOSITORY  https://github.com/rapidsai/gputreeshap.git
+            GIT_TAG         ${PKG_PINNED_TAG}
+    )
+
+    set(GPUTreeShap_ADDED ${GPUTreeShap_ADDED} PARENT_SCOPE)
+
+endfunction()
+
+find_and_configure_gputreeshap(PINNED_TAG c78fe621e429117cbca45e7b23eb5c3b6280fa3a)

cpp/include/cuml/explainer/tree_shap.hpp

@@ -0,0 +1,44 @@
+/*
+ * Copyright (c) 2021, 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
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#pragma once
+
+#include <cstddef>
+#include <cstdint>
+#include <cuml/ensemble/treelite_defs.hpp>
+#include <memory>
+
+namespace ML {
+namespace Explainer {
+
+// An abstract class representing an opaque handle to path information
+// extracted from a tree model. The implementation in tree_shap.cu will
+// define an internal class that inherits from this abtract class.
+class TreePathInfo {
+ public:
+  enum class ThresholdTypeEnum : std::uint8_t { kFloat, kDouble };
+  virtual ThresholdTypeEnum GetThresholdType() const = 0;
+};
+
+std::unique_ptr<TreePathInfo> extract_path_info(ModelHandle model);
+void gpu_treeshap(const TreePathInfo* path_info,
+                  const float* data,
+                  std::size_t n_rows,
+                  std::size_t n_cols,
+                  float* out_preds);
+
+}  // namespace Explainer
+}  // namespace ML

cpp/src/decisiontree/decisiontree.cuh

@@ -216,6 +216,7 @@ tl::Tree<T, T> build_treelite_tree(const DT::TreeMetaDataNode<T, L>& rf_tree,
           tl_tree.SetLeafVector(tl_node_id, leaf_vector);
         }
       }
+      tl_tree.SetDataCount(tl_node_id, q_node.InstanceCount());
     }
 
     cur_level_queue.swap(next_level_queue);