stepgbm (version 1.0.1)

stepgbmRVI: Select predictive variables for generalized boosted regression modeling (gbm) by relative variable influence (rvi) and accuracy in a stepwise algorithm

Description

This function is to select predictive variables for generalized boosted regression modeling (gbm) by their relative variable influence that is calculated for each model after excluding the least influence variable, and corresponding predictive accuracy. It is also developed for 'stepgbm' function.

Usage

stepgbmRVI(
  trainx,
  trainy,
  var.monotone = rep(0, ncol(trainx)),
  family = "gaussian",
  n.trees = 3000,
  learning.rate = 0.001,
  interaction.depth = 2,
  bag.fraction = 0.5,
  train.fraction = 1,
  n.minobsinnode = 10,
  cv.fold = 10,
  weights = rep(1, nrow(trainx)),
  keep.data = FALSE,
  verbose = TRUE,
  n.cores = 6,
  rpt = 2,
  predacc = "VEcv",
  min.n.var = 2,
  rseed = 1234,
  ...
)

Value

A list with the following components: variable removed based on avi (variable.removed), averaged predictive accuracy of the model after excluding variable.removed (predictive.accuracy), contribution to accuracy by each variable.removed (delta.accuracy), and predictive accuracy matrix of the model after excluding variable.removed for each iteration (predictive.accuracy2)

Arguments

trainx

a dataframe or matrix contains columns of predictive variables.

trainy

a vector of response, must have length equal to the number of rows in trainx.

var.monotone

an optional vector, the same length as the number of predictors, indicating which variables have a monotone increasing (+1), decreasing (-1), or arbitrary (0) relationship with the outcome. By default, a vector of 0 is used.

family

either a character string specifying the name of the distribution to use or a list with a component name specifying the distribution and any additional parameters needed. See gbm for details. By default, "gaussian" is used.

n.trees

the total number of trees to fit. This is equivalent to the number of iterations and the number of basis functions in the additive expansion. By default, 3000 is used.

learning.rate

a shrinkage parameter applied to each tree in the expansion. Also known as step-size reduction.

interaction.depth

the maximum depth of variable interactions. 1 implies an additive model, 2 implies a model with up to 2-way interactions, etc. By default, 2 is used.

bag.fraction

the fraction of the training set observations randomly selected to propose the next tree in the expansion. By default, 0.5 is used.

train.fraction

The first train.fraction * nrows(data) observations are used to fit the gbm and the remainder are used for computing out-of-sample estimates of the loss function.

n.minobsinnode

minimum number of observations in the trees terminal nodes. Note that this is the actual number of observations not the total weight. By default, 10 is used.

cv.fold

integer; number of cross-validation folds to perform within gbm. if > 1, then apply n-fold cross validation; the default is 10, i.e., 10-fold cross validation that is recommended.

weights

an optional vector of weights to be used in the fitting process. Must be positive but do not need to be normalized. If keep.data = FALSE in the initial call to gbm then it is the user's responsibility to resupply the weights to gbm.more. By default, a vector of 1 is used.

keep.data

a logical variable indicating whether to keep the data and an index of the data stored with the object. Keeping the data and index makes subsequent calls to gbm.more faster at the cost of storing an extra copy of the dataset. By default, 'FALSE' is used.

verbose

If TRUE, gbm will print out progress and performance indicators. By default, 'TRUE' is used.

n.cores

The number of CPU cores to use. See gbm for details. By default, 6 is used.

rpt

iteration of cross validation.

predacc

"VEcv" for vecv in function pred.acc.

min.n.var

minimum number of predictive variables remained in the final predictive model the default is 1.

rseed

random seed. By default, 1234 is used.

...

other arguments passed on to gbm.

Author

Jin Li

References

Li, J., Siwabessy, J., Huang, Z., Nichol, S. (2019). "Developing an optimal spatial predictive model for seabed sand content using machine learning, geostatistics and their hybrid methods." Geosciences 9 (4):180.

Li, J., Alvarez, B., Siwabessy, J., Tran, M., Huang, Z., Przeslawski, R., Radke, L., Howard, F., Nichol, S. (2017). "Application of random forest, generalised linear model and their hybrid methods with geostatistical techniques to count data: Predicting sponge species richness." Environmental Modelling & Software 97: 112-129.

Li, J., Alvarez, B., Siwabessy, J., Tran, M., Huang, Z., Przeslawski, R., Radke, L., Howard, F., Nichol, S. (2017). Selecting predictors to form the most accurate predictive model for count data. International Congress on Modelling and Simulation (MODSIM) 2017, Hobart.

Chang, W. 2021. Cookbook for R. http://www.cookbook-r.com/.

Examples

# \donttest{
library(spm)
data(petrel)
stepgbm1 <- stepgbmRVI(trainx = petrel[, c(1,2, 6:9)], trainy = log(petrel[, 5] + 1),
 cv.fold = 5, min.n.var = 2, n.cores = 2, rseed = 1234)
stepgbm1

#plot stepgbm1 results
library(reshape2)
pa1 <- as.data.frame(stepgbm1$predictive.accuracy2)
names(pa1) <- stepgbm1$variable.removed
pa2 <- melt(pa1, id = NULL)
names(pa2) <- c("Variable","VEcv")
library(lattice)
with(pa2, boxplot(VEcv~Variable, ylab="VEcv (%)", xlab="Predictive variable removed"))

barplot(stepgbm1$delta.accuracy, col = (1:length(stepgbm1$variable.removed)),
names.arg = stepgbm1$variable.removed, main = "Predictive accuracy vs variable removed",
font.main = 4, cex.names=1, font=2, ylab="Increase rate in VEcv (%)")

# Extract names of the selected predictive variables by stepgbm
library(steprf)
steprfAVIPredictors(stepgbm1, trainx = petrel[, c(1,2, 6:9)])

data(sponge)
set.seed(1234)
stepgbm2 <- stepgbmRVI(trainx = sponge[, -3], trainy = sponge[, 3],
family = "poisson", cv.fold = 5, min.n.var = 2, n.cores = 2)
stepgbm2

#plot stepgbm2 results
library(reshape2)
pa1 <- as.data.frame(stepgbm2$predictive.accuracy2)
names(pa1) <- stepgbm2$variable.removed
pa2 <- melt(pa1, id = NULL)
names(pa2) <- c("Variable","VEcv")
library(lattice)
with(pa2, boxplot(VEcv~Variable, ylab="VEcv (%)", xlab="Predictive variable removed"))

barplot(stepgbm2$delta.accuracy, col = (1:length(stepgbm2$variable.removed)),
names.arg = stepgbm2$variable.removed, main = "Predictive accuracy vs variable removed",
font.main = 4, cex.names=1, font=2, ylab="Increase rate in VEcv (%)")

# Extract names of the selected predictive variables by stepgbm
steprfAVIPredictors(stepgbm2, trainx =  sponge[, -3])
# }