Use consistent integer types for n_rows and row IDs in random forest

Currently, we use 32-bit int for the n_rows parameter of the random forest:

cuml/cpp/include/cuml/ensemble/randomforest.hpp

Lines 145 to 156 in 93418ba

    
           void fit(const raft::handle_t& user_handle, 
        
                    RandomForestClassifierF* forest, 
        
                    float* input, 
        
                    int n_rows, 
        
                    int n_cols, 
        
                    int* labels, 
        
                    int n_unique_labels, 
        
                    RF_params rf_params, 
        
                    rapids_logger::level_enum verbosity = rapids_logger::level_enum::info, 
        
                    bool* bootstrap_masks               = nullptr, 
        
                    const double* sample_weight         = nullptr, 
        
                    bool input_row_major                = false);

On the other hand, we use 64-bit integers to store row IDs in the internals of random forest implementation:

cuml/cpp/src/randomforest/randomforest.cuh

Lines 65 to 71 in 93418ba

    
           RowSampler(const raft::handle_t& handle, 
        
                      const RF_params& rf_params, 
        
                      std::int64_t n_rows, 
        
                      std::int64_t n_sampled_rows, 
        
                      int n_streams, 
        
                      bool* bootstrap_masks, 
        
                      const double* sample_weight)

cuml/cpp/src/randomforest/randomforest.cuh

Line 113 in 93418ba

    
           rmm::device_uvector<std::int64_t>& sample(int tree_id, int stream_id, cudaStream_t stream)

cuml/cpp/src/decisiontree/batched-levelalgo/kernels/builder_kernels.cuh

Lines 100 to 108 in 93418ba

    
           template <typename DataT, typename LabelT, int TPB> 
        
           void launchNodeSplitKernel(const Dataset<DataT, LabelT>& dataset, 
        
                                      const NodeWorkItem* work_items, 
        
                                      Split<DataT>* splits, 
        
                                      const WorkloadInfo* workload_info, 
        
                                      size_t n_blocks_dimx, 
        
                                      size_t n_work_items, 
        
                                      std::int64_t* partition_row_ids, 
        
                                      cudaStream_t builder_stream);

cuml/cpp/src/decisiontree/batched-levelalgo/kernels/builder_kernels.cuh

Lines 67 to 75 in 93418ba

    
           inline void sample_features(std::int64_t* column_samples, 
        
                                       const NodeWorkItem* work_items, 
        
                                       size_t work_items_size, 
        
                                       std::int64_t treeid, 
        
                                       uint64_t seed, 
        
                                       std::int64_t sample_offset, 
        
                                       std::int64_t n, 
        
                                       std::int64_t k, 
        
                                       cudaStream_t stream)

cuml/cpp/src/decisiontree/batched-levelalgo/builder.cuh

Lines 318 to 319 in 93418ba

    
           auto partition_row_ids_bytes = 
        
             ML::checked_mul<std::size_t>(sizeof(std::int64_t), dataset.n_sampled_rows);

TODO: We need to decide whether to use 32-bit or 64-bit integer to store n_rows and row IDs and apply the consistent type everywhere.
If we were to choose 32-bit, we'd allocate smaller memory for row IDs in the builder kernel.

TODO: Add assertion in the Python-C++ boundary to ensure that the Python integer is representable as the integer type used in the C++ implementation.

Related: #2459, #4105

Navigation Menu

Repository navigation

Use consistent integer types for `n_rows` and row IDs in random forest #8817

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

	void fit(const raft::handle_t& user_handle,
	RandomForestClassifierF* forest,
	float* input,
	int n_rows,
	int n_cols,
	int* labels,
	int n_unique_labels,
	RF_params rf_params,
	rapids_logger::level_enum verbosity = rapids_logger::level_enum::info,
	bool* bootstrap_masks = nullptr,
	const double* sample_weight = nullptr,
	bool input_row_major = false);

	RowSampler(const raft::handle_t& handle,
	const RF_params& rf_params,
	std::int64_t n_rows,
	std::int64_t n_sampled_rows,
	int n_streams,
	bool* bootstrap_masks,
	const double* sample_weight)

	template <typename DataT, typename LabelT, int TPB>
	void launchNodeSplitKernel(const Dataset<DataT, LabelT>& dataset,
	const NodeWorkItem* work_items,
	Split<DataT>* splits,
	const WorkloadInfo* workload_info,
	size_t n_blocks_dimx,
	size_t n_work_items,
	std::int64_t* partition_row_ids,
	cudaStream_t builder_stream);

	inline void sample_features(std::int64_t* column_samples,
	const NodeWorkItem* work_items,
	size_t work_items_size,
	std::int64_t treeid,
	uint64_t seed,
	std::int64_t sample_offset,
	std::int64_t n,
	std::int64_t k,
	cudaStream_t stream)

	auto partition_row_ids_bytes =
	ML::checked_mul<std::size_t>(sizeof(std::int64_t), dataset.n_sampled_rows);

Use consistent integer types for n_rows and row IDs in random forest #8817

Description

Activity

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Use consistent integer types for `n_rows` and row IDs in random forest #8817