gbm_train: Gradient boosting machine based on GPR

Description Usage Arguments Value

View source: R/boosting_gpr.R

Description

Training a gradient boosting machine

Usage

1
2
3
4
5
6
gbm_train(train_x, train_y, test_x, test_y, pred_method = "2",
  n_model = 500, batch_size = 1000, lr = 0.1, decay_lr = 1,
  tune_param = FALSE, tune_size = NULL, sr_size = NULL,
  selected_n_feature = NULL, update_kparam_tiems = 50,
  update_col_sample = 50, update_lr = 50, kname = "gaussiandotrel",
  ktheta = NULL, kbetainv = NULL, ncpu = -1)

Arguments

train_x

Matrix; the features of training data set.

train_y

Matrix; y.

pred_method

String; Set the model training approach. 1: random row sampling after all training data have been used. 2: random row sampling. 3: row sampling plus col sampling.

n_model

Positive number of submodel in gbm.

batch_size

Positive integer; batch size for each iteration.

lr

Numeric between 0-1; learning rate.

decay_lr

Numeric between 0-1; decay learning rate, default = 1 (no decay).

tune_param

Boolean; Set to TRUE to tune parameters of kernel function default value is TRUE.

tune_size

Positve integer; size of tuning data set.

sr_size

Positive integer; size of sub dataset in gbm_sr

update_col_sample

Positive integer; time to update kernel parameter(for method 3).

update_lr

Positive integer; time to decay learning rate; default is 50.

kname

String; the name of kernel; default value is 'gaussiandotrel'.

ktheta

Numeric vector; store kernel parameter; should be provided when tune_param is FALSE.

kbetainv

Numeric; store kernel parameter betainv; shuld be provided when tune_param is FALSE.

ncpu

Integer; the number of thread to be used; set to -1 to use all threads; default value is -1.

tune_param

Boolean; Set to TRUE to tune parameters of kernel function, default value for pred_method 1 & 2 is TRUE. default value for pred_method 3 is FALSE.

update_kparam_times

Positve integer; time to update kernel parameter(for method 1/2).

Value

return a list having four objects: models pred_method train_rmse test_rmse


weichunliao/baeirGPR documentation built on July 30, 2018, 10:06 a.m.