README.md

In Yael-Travis-Lumer/mlr3learners.KMforCSD: An MLR3 Regression Learner that Enables Cross-Validation for the Proposed Kernel Machine.

mlr3learners.KMforCSD

mlr3learners.KMforCSD is an R package that wraps the KMforCSD algorithm as an MLR3 learner. The mlr3learners.KMforCSD package enables cross validation with respect to the loss presented in the paper Kernel Machines for Current Status Data.

Installation

You can install the package from its GitHub repository. You first need to install the devtools package.

install.packages("devtools")

Then install mlr3learners.KMforCSD using the install_github function in the devtools package.

library(devtools)
install_github("Yael-Travis-Lumer/mlr3learners.KMforCSD")

Example

1. Initialize hyper-parameter tuner and optimization criteria:

library(mlr3)
library(mlr3tuning)
library(data.table)
library(KMforCSD)
library(mlr3learners.KMforCSD)
mlr_tuners$get("design_points")
tnr("design_points")
mlr3::mlr_measures$add("csd_risk", MeasureCSD)

1. Define train set and test set as MLR3 tasks:

n_test = 10000
n_train = 100
test_list = weibull_data(n=n_test)
test_data = test_list$data
task_test  =  TaskRegr$new(id = "Test Data", backend = test_data, target = "C")
train_list = weibull_data(n=n_train)
train_data = train_list$data
task_train =  TaskRegr$new(id = "Train Data", backend = train_data, target = "C")

1. Initializing the KM-CSD as an MLR3 learner, with either an RBF kernel or a linear kernel:

learner_rbf = lrn("regr.KMforCSD")
learner_rbf$param_set$values$kernel = "rbfdot"
learner_linear = lrn("regr.KMforCSD")
learner_linear$param_set$values$kernel = "vanilladot"

1. Use cross-validation: ```{r} cv_cost_val=c(0.1,1,10,100) # cost CV values cv_gamma_val=c(3,12,50,200) # RBF width CV values

Tuning hyper-parameters

terminator_linear = trm("evals", n_evals = length(cv_cost_val)) terminator_rbf = trm("evals", n_evals = length(cv_cost_val)*length(cv_gamma_val))

CV for rbf kernel

tune_ps = ParamSet$new(list( ParamDbl$new("cost", lower = min(cv_cost_val), upper = max(cv_cost_val)), ParamDbl$new("gamma", lower = min(cv_gamma_val), upper = max(cv_gamma_val)) )) instance_rbf = TuningInstanceSingleCrit$new( task = task_train, learner = learner_rbf, resampling = rsmp("cv", folds = 5), measure = msr("csd_risk"), search_space = tune_ps, terminator = terminator_rbf ) design = data.table(cost=sort(rep(cv_cost_val,times=length(cv_gamma_val))),gamma=rep(cv_gamma_val,length(cv_cost_val))) tuner = tnr("design_points", design = design) result = tuner$optimize(instance_rbf)

Add best lambda to learner

learner_rbf$param_set$values = instance_rbf$result_learner_param_vals

train one last time with best lambda

trained_rbf = learner_rbf$train(task_train) predictions_rbf = learner_rbf$predict(task_train) Expectation_rbf=predictions_rbf$response print("completed RBF training")

testing

predictions_rbf_test = learner_rbf$predict(task_test) Expectation_rbf_test=predictions_rbf_test$response print("completed RBF testing")

CV for linear kernel

tune_ps = ParamSet$new(list( ParamDbl$new("cost", lower = min(cv_cost_val), upper = max(cv_cost_val)) )) instance_linear = TuningInstanceSingleCrit$new( task = task_train, learner = learner_linear, resampling = rsmp("cv", folds = 5), measure = msr("csd_risk"), search_space = tune_ps, terminator = terminator_linear ) design2 = data.table(cost=cv_cost_val) tuner = tnr("design_points", design = design2) result = tuner$optimize(instance_linear)

#Add best lambda to learner

learner_linear$param_set$values = instance_linear$result_learner_param_vals

train one last time with best lambda

trained_linear = learner_linear$train(task_train) predictions_linear = learner_linear$predict(task_train) Expectation_linear=predictions_linear$response print("completed linear training")

testing

predictions_linear_test = learner_linear$predict(task_test) Expectation_linear_test=predictions_linear_test$response print("completed linear testing")

Yael-Travis-Lumer/mlr3learners.KMforCSD documentation built on Sept. 20, 2020, 11:49 a.m.