secretflow.ml.boost.ss_xgb_v.core

secretflow.ml.boost.ss_xgb_v.core.node_split

Classes:

RegType(value)

An enumeration.

Functions:

sigmoid(pred)
compute_obj(G, H, reg_lambda)	compute objective values of input buckets.
compute_weight(G, H, reg_lambda, learning_rate)	compute weight values of tree leaf nodes.
compute_gh(y, pred, objective)	compute first and second order gradient of each sample.
tree_setup(pred, y, sub_choices, objective)	Set up pre-tree context.
compute_gradient_sums(nodes_s, cache, ...)
find_best_split_bucket(GHs, reg_lambda)	compute the gradient sums of the containing instances in each split bucket and find best split bucket for each node which has the max split gain.
init_pred(base, samples)
root_select(samples)
get_child_select(nodes_s, lchilds_ss, fragments)	compute the next level's select indexes.
predict_tree_weight(selects, weights)	get final pred for this tree.
get_weight(sums, reg_lambda, learning_rate)
sum_leaf(ss, gh, sub_choices)
update_train_pred(pred, current, fragments)

class secretflow.ml.boost.ss_xgb_v.core.node_split.RegType(value)

基类：Enum

An enumeration.

Attributes:

Linear
Logistic

Linear = 'linear'

Logistic = 'logistic'

secretflow.ml.boost.ss_xgb_v.core.node_split.sigmoid(pred: ndarray) → ndarray

secretflow.ml.boost.ss_xgb_v.core.node_split.compute_obj(G: ndarray, H: ndarray, reg_lambda: float) → ndarray

compute objective values of input buckets.

参数:

• G/H – sum of first and second order gradient in each bucket.

• reg_lambda – L2 regularization term

返回:

objective values.

secretflow.ml.boost.ss_xgb_v.core.node_split.compute_weight(G: float, H: float, reg_lambda: float, learning_rate: float) → ndarray

compute weight values of tree leaf nodes.

参数:

• G/H – sum of first and second order gradient in each node.

• reg_lambda – L2 regularization term

• learning_rate – Step size shrinkage used in update to prevents overfitting.

返回:

weight values.

secretflow.ml.boost.ss_xgb_v.core.node_split.compute_gh(y: ndarray, pred: ndarray, objective: RegType) → Tuple[ndarray, ndarray]

compute first and second order gradient of each sample.

参数:

• y – sample true label of each sample.

• pred – prediction of each sample.

• objective – regression learning objective,

返回:

weight values.

secretflow.ml.boost.ss_xgb_v.core.node_split.tree_setup(pred: ndarray, y: ndarray, sub_choices: ndarray, objective: RegType) → Tuple[ndarray, ndarray]

Set up pre-tree context.

secretflow.ml.boost.ss_xgb_v.core.node_split.compute_gradient_sums(nodes_s: List[ndarray], cache: List[List[ndarray]], col_choices: ndarray, sub_choices: ndarray, gh: List[ndarray], buckets_map: ndarray)

secretflow.ml.boost.ss_xgb_v.core.node_split.find_best_split_bucket(GHs: List[List[ndarray]], reg_lambda: float) → Tuple[ndarray, Dict[str, Any]]

compute the gradient sums of the containing instances in each split bucket and find best split bucket for each node which has the max split gain.

参数:

• context – comparison context.

• nodes_s – sample select indexes of each node from same tree level.

• last_level – if this split is last level, next level is leaf nodes.

返回:

idx of split bucket for each node.

secretflow.ml.boost.ss_xgb_v.core.node_split.init_pred(base: float, samples: int)

secretflow.ml.boost.ss_xgb_v.core.node_split.root_select(samples: int) → List[ndarray]

secretflow.ml.boost.ss_xgb_v.core.node_split.get_child_select(nodes_s: List[ndarray], lchilds_ss: List[ndarray], fragments: int) → List[ndarray]

compute the next level’s select indexes.

参数:

• nodes_s – sample select indexes of each node from current level’s nodes.

• lchilds_ss – left children’s sample select idx for current level’s nodes.

返回:

sample select indexes for nodes in next tree level.

secretflow.ml.boost.ss_xgb_v.core.node_split.predict_tree_weight(selects: List[ndarray], weights: ndarray) → ndarray

get final pred for this tree.

参数:

• selects – leaf nodes’ sample selects from each model handler.

• weights – leaf weights in secure share.

返回:

pred

secretflow.ml.boost.ss_xgb_v.core.node_split.get_weight(sums: List[List[ndarray]], reg_lambda: float, learning_rate: float) → ndarray

secretflow.ml.boost.ss_xgb_v.core.node_split.sum_leaf(ss: List[ndarray], gh: List[ndarray], sub_choices: ndarray)

secretflow.ml.boost.ss_xgb_v.core.node_split.update_train_pred(pred: List[ndarray], current: ndarray, fragments: int)

secretflow.ml.boost.ss_xgb_v.core.tree_worker

Classes:

XgbTreeWorker

ActorProxy(XgbTreeWorker) 的别名

secretflow.ml.boost.ss_xgb_v.core.tree_worker.XgbTreeWorker

ActorProxy(XgbTreeWorker) 的别名 Methods:

__init__(*args, **kwargs)	Abstraction device object base class.
predict_weight_select(x, tree)	computer leaf nodes' sample selects known by this partition.
build_bucket_map(start, length)	Build bucket_map fragment base on order_map.
global_setup(x, buckets, seed)	Set up global context.
update_buckets_count(buckets_count, ...)	save how many buckets in each partition's features.
tree_setup(colsample)	Set up tree context and do col sample if colsample < 1
tree_finish()
do_split(split_buckets)	record split info and generate next level's left children select.

secretflow.ml.boost.ss_xgb_v.core.utils

Functions:

prepare_dataset(ds)

check data setting and get total shape.

secretflow.ml.boost.ss_xgb_v.core.utils.prepare_dataset(ds: Union[FedNdarray, VDataFrame]) → Tuple[FedNdarray, Tuple[int, int]]

check data setting and get total shape.

参数:

ds – input dataset

返回:

dataset in unified type Second: shape concat all partition.

返回类型:

First

secretflow.ml.boost.ss_xgb_v.core.xgb_tree

Classes:

XgbTree()

class secretflow.ml.boost.ss_xgb_v.core.xgb_tree.XgbTree

基类：object

Methods:

__init__()
insert_split_node(feature, value)

__init__() → None

insert_split_node(feature: int, value: float) → None