rbooster
This package aims to enable the use of classifiers in boosting family algorithms, apart from decision trees and other classical weak classifiers. The current scope of the package includes Real AdaBoost and Discrete AdaBoost, as well as multiclass versions of these methods, the SAMME and SAMME.R algorithms.
I am aware that there have been attempts to create frameworks that allow using custom functions in boosting (e.g. boostr). However, they are too complex to my taste. I haven’t seen a simple-to-use package that handles different boosting methods. What can be done simply in Python, why can’t it be done in R? That is my point. The package is still in the starting phase (a baby package). Things on my mind:
booster
functionThere is a function for each method in the package. But the main function is the booster
. By using this function it will be possible to access everything. Let me give you information about the features of this function and how to use it. First of all, 2 different AdaBoost methods can be applied with the booster
function: Real AdaBoost and Discrete AdaBoost. Let us use glass0 dataset in imbalance
package and prepare a train and test dataset. test dataset is not mandatory.
library(rbooster)
<- function(y, train_proportion) {
cv_sampler unlist(lapply(unique(y), function(m) sample(which(y==m), round(sum(y==m))*train_proportion)))
}
library(imbalance)
<- glass0
data <- ncol(data) - 1
p
<- data[,1:p]
x <- data[, p + 1]
y
<- cv_sampler(y, 0.9)
train_i <- x[train_i,]
x_train <- y[train_i]
y_train
<- x[-train_i,]
x_test <- y[-train_i] y_test
This is a two-class dataset. Let us use decision tree to build Discrete and Real AdaBoost models.
<- booster(x_train = x_train,
m_discrete y_train = y_train,
classifier = "rpart",
method = "discrete",
x_test = x_test,
y_test = y_test,
weighted_bootstrap = FALSE,
max_iter = 20,
lambda = 1,
print_detail = TRUE,
print_plot = TRUE,
bag_frac = 0.8,
p_weak = 4)
#> 1 Train err:0.04167, Test err:0.13636, Weighted err:0.04167
#> 2 Train err:0.04167, Test err:0.13636, Weighted err:0.08544
#> 3 Train err:0.02604, Test err:0.18182, Weighted err:0.10124
#> 4 Train err:0.00521, Test err:0.13636, Weighted err:0.08556
#> 5 Train err:0.00521, Test err:0.09091, Weighted err:0.08887
#> 6 Train err:0.00521, Test err:0.13636, Weighted err:0.15156
#> 7 Train err:0.00521, Test err:0.09091, Weighted err:0.22742
#> 8 Train err:0.00521, Test err:0.13636, Weighted err:0.11090
#> 9 Train err:0.00000, Test err:0.13636, Weighted err:0.05789
#> 10 Train err:0.00000, Test err:0.18182, Weighted err:0.14545
#> 11 Train err:0.00000, Test err:0.13636, Weighted err:0.13199
#> 12 Train err:0.00000, Test err:0.18182, Weighted err:0.08357
#> 13 Train err:0.00000, Test err:0.09091, Weighted err:0.10590
#> 14 Train err:0.00000, Test err:0.09091, Weighted err:0.06426
#> 15 Train err:0.00000, Test err:0.09091, Weighted err:0.13477
#> 16 Train err:0.00000, Test err:0.04545, Weighted err:0.22908
#> 17 Train err:0.00000, Test err:0.00000, Weighted err:0.14548
#> 18 Train err:0.00000, Test err:0.04545, Weighted err:0.16395
#> 19 Train err:0.00000, Test err:0.00000, Weighted err:0.11084
#> 20 Train err:0.00000, Test err:0.04545, Weighted err:0.10292
<- booster(x_train = x_train,
m_real y_train = y_train,
classifier = "rpart",
method = "real",
x_test = x_test,
y_test = y_test,
weighted_bootstrap = FALSE,
max_iter = 20,
lambda = 1,
print_detail = TRUE,
print_plot = TRUE,
bag_frac = 0.8,
p_weak = 4)
#> 1 Train err:0.09896, Test err:0.31818, Weighted err:0.09896
#> 2 Train err:0.01562, Test err:0.31818, Weighted err:0.02068
#> 3 Train err:0.01562, Test err:0.27273, Weighted err:0.03289
#> 4 Train err:0.01562, Test err:0.18182, Weighted err:0.11748
#> 5 Train err:0.01042, Test err:0.18182, Weighted err:0.03072
#> 6 Train err:0.00521, Test err:0.22727, Weighted err:0.04445
#> 7 Train err:0.00521, Test err:0.13636, Weighted err:0.04787
#> 8 Train err:0.00000, Test err:0.22727, Weighted err:0.11679
#> 9 Train err:0.00000, Test err:0.18182, Weighted err:0.04760
#> 10 Train err:0.01042, Test err:0.18182, Weighted err:0.24894
#> 11 Train err:0.00000, Test err:0.13636, Weighted err:0.11990
#> 12 Train err:0.00000, Test err:0.18182, Weighted err:0.01676
#> 13 Train err:0.00000, Test err:0.18182, Weighted err:0.04454
#> 14 Train err:0.00000, Test err:0.22727, Weighted err:0.11525
#> 15 Train err:0.00000, Test err:0.13636, Weighted err:0.00707
#> 16 Train err:0.00000, Test err:0.18182, Weighted err:0.02684
#> 17 Train err:0.00000, Test err:0.13636, Weighted err:0.23579
#> 18 Train err:0.00000, Test err:0.13636, Weighted err:0.21639
#> 19 Train err:0.00000, Test err:0.13636, Weighted err:0.73959
#> Warning in bb(x_train = x_train, y_train = y_train, classifier = classifier, :
#> Learning stopped at iteration 19
We can obtain predictions on x_test
like:
head(m_discrete$test_prediction)
#> [1] "positive" "positive" "positive" "positive" "positive" "positive"
head(m_real$test_prediction)
#> [1] "positive" "positive" "positive" "positive" "positive" "negative"
table(y_test, m_discrete$test_prediction)
#>
#> y_test negative positive
#> negative 14 1
#> positive 0 7
table(y_test, m_real$test_prediction)
#>
#> y_test negative positive
#> negative 13 2
#> positive 1 6
However x_test
is used to validate on y_test
. So without y_test
, x_test
is ignored. x_test
and y_test
can be set to NULL
, they are not mandatory for model to be established. You can use the usual predict
function.
<- predict(object = m_discrete, newdata = x_test, type = "pred")
pred_discrete <- predict(object = m_real, newdata = x_test, type = "pred")
pred_real
all(pred_discrete == m_discrete$test_prediction)
#> [1] TRUE
all(pred_discrete == m_discrete$test_prediction)
#> [1] TRUE
In predict
function type
can be “pred” or “prob”. “prob” will gives class probability matrix.
<- predict(object = m_discrete, newdata = x_test, type = "prob")
prob_discrete head(prob_discrete)
#> negative positive
#> [1,] 4.439571e-06 0.9999956
#> [2,] 3.711468e-03 0.9962885
#> [3,] 4.418085e-08 1.0000000
#> [4,] 5.589519e-05 0.9999441
#> [5,] 2.483800e-02 0.9751620
#> [6,] 3.266116e-02 0.9673388
Let us use Glass dataset in mlbench for multiclass classification example.
library(mlbench)
data(Glass)
<- Glass
data <- ncol(data) - 1
p
<- data[,1:p]
x <- data[, p + 1]
y
<- cv_sampler(y, 0.9)
train_i <- x[train_i,]
x_train <- y[train_i]
y_train
<- x[-train_i,]
x_test <- y[-train_i]
y_test
par(mfrow = c(2,1))
<- booster(x_train = x_train,
m_discrete y_train = y_train,
classifier = "rpart",
method = "discrete",
x_test = x_test,
y_test = y_test,
weighted_bootstrap = FALSE,
max_iter = 20,
lambda = 1,
print_detail = FALSE,
print_plot = TRUE,
bag_frac = 0.8,
p_weak = p)
<- booster(x_train = x_train,
m_real y_train = y_train,
classifier = "rpart",
method = "real",
x_test = x_test,
y_test = y_test,
weighted_bootstrap = FALSE,
max_iter = 20,
lambda = 0.1,
print_detail = FALSE,
print_plot = TRUE,
bag_frac = 1,
p_weak = p)
#> Warning in bb(x_train = x_train, y_train = y_train, classifier = classifier, :
#> Learning stopped at iteration 12
invisible(dev.off())
<- predict(object = m_discrete, newdata = x_test, type = "pred")
pred_discrete <- predict(object = m_real, newdata = x_test, type = "pred")
pred_real
table(y_test, pred_discrete)
#> pred_discrete
#> y_test 1 2 3 5 6 7
#> 1 4 3 0 0 0 0
#> 2 1 6 0 0 1 0
#> 3 0 1 1 0 0 0
#> 5 0 1 0 0 0 1
#> 6 0 0 0 0 1 0
#> 7 0 0 0 0 0 3
table(y_test, pred_real)
#> pred_real
#> y_test 1 2 3 5 6 7
#> 1 4 2 1 0 0 0
#> 2 0 5 2 0 1 0
#> 3 0 0 2 0 0 0
#> 5 0 0 0 1 0 1
#> 6 1 0 0 0 0 0
#> 7 0 0 0 0 0 3
There is a major reason people use decision trees in boosting. It fits very quick. Let us try a pre-ready classifier Discrete Naive Bayes, “dnb”, in rbooster
.
par(mfrow = c(2,1))
<- booster(x_train = x_train,
m_discrete y_train = y_train,
classifier = "dnb",
method = "discrete",
x_test = x_test,
y_test = y_test,
weighted_bootstrap = FALSE,
max_iter = 250,
lambda = 1,
print_detail = FALSE,
print_plot = TRUE,
bag_frac = 0.5,
p_weak = 4)
<- booster(x_train = x_train,
m_real y_train = y_train,
classifier = "dnb",
method = "real",
x_test = x_test,
y_test = y_test,
weighted_bootstrap = FALSE,
max_iter = 250,
lambda = 1e-4,
print_detail = FALSE,
print_plot = TRUE,
bag_frac = 0.2,
p_weak = 4)
invisible(dev.off())
<- predict(object = m_discrete, newdata = x_test, type = "pred")
pred_discrete <- predict(object = m_real, newdata = x_test, type = "pred")
pred_real
table(y_test, pred_discrete)
#> pred_discrete
#> y_test 1 2 3 5 6 7
#> 1 2 4 1 0 0 0
#> 2 0 8 0 0 0 0
#> 3 1 0 1 0 0 0
#> 5 0 0 0 1 0 1
#> 6 0 1 0 0 0 0
#> 7 0 1 0 0 0 2
table(y_test, pred_real)
#> pred_real
#> y_test 1 2 3 5 6 7
#> 1 3 4 0 0 0 0
#> 2 0 7 0 0 1 0
#> 3 2 0 0 0 0 0
#> 5 0 0 0 1 0 1
#> 6 0 0 0 0 0 1
#> 7 0 0 0 0 0 3
I needed to play with lambda, bag_frac and p_weak a lot to find a good setting so that “dnb” fits. It is still not quite successful. Methods other that decision trees require care.
Classifier function requires x_train, y_train, weights as mandatory. Output of Classifier function will be the input for Predictor function. Predictor function requires model, which is the output of Classifier, x_new, which is the train features, and type inputs. type can only be “pred” or “prob”. “pred” must give a class vector of predictions, and “prob” must give a matrix of class probabilities. “pred” is mandatory for both Real and Discrete AdaBoost. However, Discrete AdaBoost does not require type “prob”. Let me prepare a made up Classifier function.
<- function(x_train, y_train, weights, ...){
classifier_lm <- c(-1,1)
y_train_code <- sapply(levels(y_train), function(m) y_train_code[(y_train == m) + 1])
y_train_coded <- y_train_coded[,1]
y_train_coded if (is.null(weights)) {
<- rep(1, length(y_train))
weights
}
<- lm.wfit(x = as.matrix(cbind(1,x_train)), y = y_train_coded, w = weights)
model return(list(coefficients = model$coefficients,
levels = levels(y_train)))
}
<- function(model, x_new, type = "pred", ...) {
predictor_lm <- model$coefficients
coef <- model$levels
levels
<- as.matrix(cbind(1, x_new))%*%coef
fit <- 1/(1 + exp(-fit))
probs <- data.frame(probs, 1 - probs)
probs colnames(probs) <- levels
if (type == "pred") {
<- factor(levels[apply(probs, 1, which.max)], levels = levels, labels = levels)
preds return(preds)
}if (type == "prob") {
return(probs)
} }
This classifier will convert y_train into (-1,1) codes and predictions will be transformed into the range (0,1) using logit transformation. These values will be used as probabilities and class with the higher probabilitiy will be selected as prediction. Since this is a linear classification method, it should work well with boosting. Let us see.
<- glass0
data <- ncol(data) - 1
p
<- data[,1:p]
x <- data[, p + 1]
y
<- cv_sampler(y, 0.9)
train_i <- x[train_i,]
x_train <- y[train_i]
y_train
<- x[-train_i,]
x_test <- y[-train_i]
y_test
par(mfrow = c(2,1))
<- booster(x_train = x_train,
m_discrete y_train = y_train,
classifier = classifier_lm,
predictor = predictor_lm,
method = "discrete",
x_test = x_test,
y_test = y_test,
weighted_bootstrap = FALSE,
max_iter = 600,
lambda = 2,
print_detail = FALSE,
print_plot = TRUE,
bag_frac = 0.4,
p_weak = 4)
<- booster(x_train = x_train,
m_real y_train = y_train,
classifier = classifier_lm,
predictor = predictor_lm,
method = "real",
x_test = x_test,
y_test = y_test,
weighted_bootstrap = FALSE,
max_iter = 200,
lambda = 0.1,
print_detail = FALSE,
print_plot = TRUE,
bag_frac = 1,
p_weak = 4)
#> Warning in bb(x_train = x_train, y_train = y_train, classifier = classifier, :
#> Learning stopped at iteration 104
invisible(dev.off())
<- predict(object = m_discrete, newdata = x_test, type = "pred")
pred_discrete <- predict(object = m_real, newdata = x_test, type = "pred")
pred_real
table(y_test, pred_discrete)
#> pred_discrete
#> y_test negative positive
#> negative 14 1
#> positive 4 3
table(y_test, pred_real)
#> pred_real
#> y_test negative positive
#> negative 14 1
#> positive 6 1
It fits but not quite there. print_plot
can be set to FALSE
. In that case, the same plot can be generated using plot
function.
par(mfrow = c(2,1))
plot(m_discrete)
plot(m_real)
invisible(dev.off())
Pre-ready weak classifiers are “rpart”, “gnb”, “dnb”, “earth” and “glm” which are CART, Gaussian Naive Bayes, Discrete Naive Bayes, Multivariate Adaptive Regression Splines from earth package and logistic regression respectively. “glm” is not capable of multiclass classification. booster object can be used with prediction, plot, and print functions. For other details about booster object and its outputs, please see help document.