Skip to content

Commit

Permalink
[REVIEW] Refactoring: move internal FIL interface to a separate file (#…
Browse files Browse the repository at this point in the history 
…3292)

* Refactoring: move internal FIL interface to a separate file.

- move the functions not related to treelite import, prediction
  or freeing the model to a separate file

* Fixed style errors.
  • Loading branch information
@canonizer
canonizer authored Dec 14, 2020
1 parent 9fa7112 commit 62e2152
Show file tree
Hide file tree
Showing 4 changed files with 230 additions and 200 deletions.
200 changes: 1 addition & 199 deletions cpp/include/cuml/fil/fil.h
View file
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2019, NVIDIA CORPORATION.
* Copyright (c) 2019-2020, NVIDIA CORPORATION.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
Expand Down Expand Up @@ -46,36 +46,6 @@ enum algo_t {
BATCH_TREE_REORG
};

/**
* output_t are flags that define the output produced by the FIL predictor; a
* valid output_t values consists of the following, combined using '|' (bitwise
* or), which define stages, which operation in the next stage applied to the
* output of the previous stage:
* - one of RAW or AVG, indicating how to combine individual tree outputs into the forest output
* - optional SIGMOID for applying the sigmoid transform
* - optional CLASS, to output the class label
*/
enum output_t {
/** raw output: the sum of the tree outputs; use for GBM models for
regression, or for binary classification for the value before the
transformation; note that this value is 0, and may be omitted
when combined with other flags */
RAW = 0x0,
/** average output: divide the sum of the tree outputs by the number of trees
before further transformations; use for random forests for regression
and binary classification for the probability */
AVG = 0x1,
/** sigmoid transformation: apply 1/(1+exp(-x)) to the sum or average of tree
outputs; use for GBM binary classification models for probability */
SIGMOID = 0x10,
/** output class label: either apply threshold to the output of the previous stage (for binary classification),
or select the class with the most votes to get the class label (for multi-class classification). */
CLASS = 0x100,
SIGMOID_CLASS = SIGMOID | CLASS,
AVG_CLASS = AVG | CLASS,
AVG_SIGMOID_CLASS = AVG | SIGMOID | CLASS,
};

/** storage_type_t defines whether to import the forests as dense or sparse */
enum storage_type_t {
/** decide automatically; currently always builds dense forests */
Expand All @@ -92,147 +62,11 @@ enum storage_type_t {
SPARSE8,
};

/** val_t is the payload within a FIL leaf */
union val_t {
/** threshold value for branch node or output value (e.g. class
probability or regression summand) for leaf node */
float f;
/** class label */
int idx;
};

/** dense_node_t is a node in a densely-stored forest */
struct dense_node_t {
val_t val;
int bits;
};

/** sparse_node16_extra_data is what's missing from a dense node to store
a sparse node, that is, extra indexing information due to compressing
a sparse tree. */
struct sparse_node16_extra_data {
int left_idx;
int dummy; // make alignment explicit and reserve for future use
};

/** sparse_node16_t is a 16-byte node in a sparsely-stored forest */
struct sparse_node16_t : dense_node_t, sparse_node16_extra_data {
sparse_node16_t() = default;
sparse_node16_t(dense_node_t dn, sparse_node16_extra_data ed)
: dense_node_t(dn), sparse_node16_extra_data(ed) {}
};

/** sparse_node8_t is a node of reduced size (8 bytes)
in a sparsely-stored forest */
struct sparse_node8_t : dense_node_t {
sparse_node8_t() = default;
sparse_node8_t(dense_node_t dn) : dense_node_t(dn) {}
};

/** leaf_algo_t describes what the leaves in a FIL forest store (predict)
and how FIL aggregates them into class margins/regression result/best class
**/
enum leaf_algo_t {
/** storing a class probability or regression summand. We add all margins
together and determine regression result or use threshold to determine
one of the two classes. **/
FLOAT_UNARY_BINARY = 0,
/** storing a class label. Trees vote on the resulting class.
Probabilities are just normalized votes. */
CATEGORICAL_LEAF = 1,
/** 1-vs-rest, or tree-per-class, where trees are assigned round-robin to
consecutive categories and predict a floating-point margin. Used in
Gradient Boosted Decision Trees. We sum margins for each group separately
**/
GROVE_PER_CLASS = 2,
/** 1-vs-rest, or tree-per-class, where trees are assigned round-robin to
consecutive categories and predict a floating-point margin. Used in
Gradient Boosted Decision Trees. We sum margins for each group separately
This is a more specific version of GROVE_PER_CLASS.
_FEW_CLASSES means fewer (or as many) classes than threads. **/
GROVE_PER_CLASS_FEW_CLASSES = 3,
/** 1-vs-rest, or tree-per-class, where trees are assigned round-robin to
consecutive categories and predict a floating-point margin. Used in
Gradient Boosted Decision Trees. We sum margins for each group separately
This is a more specific version of GROVE_PER_CLASS.
_MANY_CLASSES means more classes than threads. **/
GROVE_PER_CLASS_MANY_CLASSES = 4,
// to be extended
};

template <leaf_algo_t leaf_algo>
struct leaf_output_t {};
template <>
struct leaf_output_t<leaf_algo_t::FLOAT_UNARY_BINARY> {
typedef float T;
};
template <>
struct leaf_output_t<leaf_algo_t::CATEGORICAL_LEAF> {
typedef int T;
};
template <>
struct leaf_output_t<leaf_algo_t::GROVE_PER_CLASS_FEW_CLASSES> {
typedef float T;
};
template <>
struct leaf_output_t<leaf_algo_t::GROVE_PER_CLASS_MANY_CLASSES> {
typedef float T;
};

/** node_init initializes node from paramters */
void node_init(dense_node_t* n, val_t output, float thresh, int fid,
bool def_left, bool is_leaf);
void node_init(sparse_node16_t* node, val_t output, float thresh, int fid,
bool def_left, bool is_leaf, int left_index);
void node_init(sparse_node8_t* node, val_t output, float thresh, int fid,
bool def_left, bool is_leaf, int left_index);

/** node_decode extracts individual members from node */
void node_decode(const dense_node_t* node, val_t* output, float* thresh,
int* fid, bool* def_left, bool* is_leaf);
void node_decode(const sparse_node16_t* node, val_t* output, float* thresh,
int* fid, bool* def_left, bool* is_leaf, int* left_index);
void node_decode(const sparse_node8_t* node, val_t* output, float* thresh,
int* fid, bool* def_left, bool* is_leaf, int* left_index);

struct forest;

/** forest_t is the predictor handle */
typedef forest* forest_t;

/** forest_params_t are the trees to initialize the predictor */
struct forest_params_t {
// total number of nodes; ignored for dense forests
int num_nodes;
// maximum depth; ignored for sparse forests
int depth;
// ntrees is the number of trees
int num_trees;
// num_cols is the number of columns in the data
int num_cols;
// leaf_algo determines what the leaves store (predict)
leaf_algo_t leaf_algo;
// algo is the inference algorithm;
// sparse forests do not distinguish between NAIVE and TREE_REORG
algo_t algo;
// output is the desired output type
output_t output;
// threshold is used to for classification if leaf_algo == FLOAT_UNARY_BINARY && (output & OUTPUT_CLASS) != 0 && !predict_proba,
// and is ignored otherwise
float threshold;
// global_bias is added to the sum of tree predictions
// (after averaging, if it is used, but before any further transformations)
float global_bias;
// only used for CATEGORICAL_LEAF inference. since we're storing the
// labels in leaves instead of the whole vector, this keeps track
// of the number of classes
int num_classes;
// blocks_per_sm, if nonzero, works as a limit to improve cache hit rate for larger forests
// suggested values (if nonzero) are from 2 to 7
// if zero, launches ceildiv(num_rows, NITEMS) blocks
int blocks_per_sm;
};

/** treelite_params_t are parameters for importing treelite models */
struct treelite_params_t {
// algo is the inference algorithm
Expand All @@ -253,38 +87,6 @@ struct treelite_params_t {
int blocks_per_sm;
};

/** init_dense uses params and nodes to initialize the dense forest stored in pf
* @param h cuML handle used by this function
* @param pf pointer to where to store the newly created forest
* @param nodes nodes for the forest, of length
(2**(params->depth + 1) - 1) * params->ntrees
* @param params pointer to parameters used to initialize the forest
*/
void init_dense(const raft::handle_t& h, forest_t* pf,
const dense_node_t* nodes, const forest_params_t* params);

/** init_sparse uses params, trees and nodes to initialize the sparse forest
* with 16-byte nodes stored in pf
* @param h cuML handle used by this function
* @param pf pointer to where to store the newly created forest
* @param trees indices of tree roots in the nodes arrray, of length params->ntrees
* @param nodes nodes for the forest, of length params->num_nodes
* @param params pointer to parameters used to initialize the forest
*/
void init_sparse(const raft::handle_t& h, forest_t* pf, const int* trees,
const sparse_node16_t* nodes, const forest_params_t* params);

/** init_sparse uses params, trees and nodes to initialize the sparse forest
* with 8-byte nodes stored in pf
* @param h cuML handle used by this function
* @param pf pointer to where to store the newly created forest
* @param trees indices of tree roots in the nodes arrray, of length params->ntrees
* @param nodes nodes for the forest, of length params->num_nodes
* @param params pointer to parameters used to initialize the forest
*/
void init_sparse(const raft::handle_t& h, forest_t* pf, const int* trees,
const sparse_node8_t* nodes, const forest_params_t* params);

/** from_treelite uses a treelite model to initialize the forest
* @param handle cuML handle used by this function
* @param pforest pointer to where to store the newly created forest
Expand Down
3 changes: 2 additions & 1 deletion cpp/src/fil/common.cuh
View file
Original file line number Diff line number Diff line change
Expand Up @@ -23,9 +23,10 @@
#include <stdexcept>
#include <string>

#include <cuml/fil/fil.h>
#include <raft/cuda_utils.cuh>

#include <cuml/fil/fil.h>
#include "internal.cuh"

namespace ML {
namespace fil {
Expand Down
Loading

0 comments on commit 62e2152

Please sign in to comment.