anibalsilva1 / modeloptimizationir

Model Optimization in Imbalanced Regression using SERA

ModelOptimizationIR

Installation

You can install the development version of ModelOptimizationIR like so:

install(devtools)
devtools::install_github("anibalsilva1/ModelOptimizationIR")

Example

Usage example of SERA as a loss function to infer extreme values:

library(ModelOptimizationIR)
library(xgboost)
library(dplyr)
library(IRon)

data("NO2Emissions")
n <- nrow(NO2Emissions)
s <- sample(1:n, size = n*0.8)

I = 1000
steps <- seq(0, 1, 1/I)

formula <- LNO2 ~ .
train <- NO2Emissions %>% dplyr::slice(s)
test <- NO2Emissions %>% dplyr::slice(-s)

y <- train$LNO2

control.points <- matrix(c(1.1, 0, 0, 3.7, 0, 0, 5, 1, 0), byrow = TRUE, ncol=3)
ph.ctrl <- phi.control(y = y, method = "range", control.pts = control.points)
phi <- phi(y = y, phi.parms = ph.ctrl)
sigma <- sigma(phis = phi, steps = steps)

# If you want to add some model parameters
params <- list(max_depth=5, eta = 10^{-1})

res <- XGBoost.sera(formula = formula,
                    train = train,
                    test = test,
                    sigma = sigma,
                    parameters = params)
res
#> $trues
#>   [1] 3.31419 4.34640 3.23080 3.67377 3.65584 3.93183 3.77963 3.75420 3.45632
#>  [10] 3.94546 3.48431 4.13035 4.01998 2.62467 4.23120 1.33500 4.58190 4.39198
#>  [19] 3.01062 2.64617 4.62399 3.02529 3.66356 3.48431 4.30271 3.85862 3.84802
#>  [28] 4.02535 4.21656 2.44235 4.07414 4.02356 4.25135 4.41280 3.08649 3.10459
#>  [37] 4.25561 4.27528 2.78501 4.36055 2.79728 3.59182 3.02042 4.24850 3.21487
#>  [46] 3.36730 4.03954 4.14630 4.07073 3.91202 4.28496 3.15274 4.40305 3.34990
#>  [55] 4.49647 3.54385 3.17388 4.22244 4.23700 3.31419 4.18662 3.81331 5.02651
#>  [64] 3.72569 4.31348 3.48431 3.88156 2.42480 3.71601 3.27714 4.10594 3.28466
#>  [73] 3.81551 3.97218 4.92435 2.89037 3.32504 3.79098 3.70130 4.57883 2.95491
#>  [82] 3.37417 4.76217 4.01458 1.36098 4.40916 2.85071 3.92197 4.09101 3.99268
#>  [91] 3.66099 3.65584 4.02535 4.21804 4.47050 3.44042 2.98062 3.99083 3.98155
#> [100] 2.95491
#> 
#> $preds
#>   [1] 3.899232 4.163350 3.871713 4.228779 4.263709 4.768865 4.194416 4.013278
#>   [9] 4.135949 3.823129 4.290694 4.197311 4.285467 4.086702 4.140828 3.859711
#>  [17] 4.353094 4.237541 4.261664 3.871822 4.680878 4.518871 4.207847 4.238189
#>  [25] 4.232188 4.139825 4.195663 4.895604 5.097433 4.016120 4.281458 4.366767
#>  [33] 4.260844 4.169632 3.910839 4.169408 4.179049 4.563083 4.137344 4.400470
#>  [41] 3.836796 3.961806 4.531759 4.680306 4.083981 4.339976 4.244241 4.498428
#>  [49] 4.692109 4.092527 4.251806 3.792355 4.253544 4.097158 4.334836 4.162409
#>  [57] 4.196109 4.097247 4.198833 3.973488 4.214164 3.818469 4.475795 4.117335
#>  [65] 4.206036 4.033762 4.311665 4.023857 4.129992 3.822489 4.070420 4.297572
#>  [73] 4.097683 4.222573 4.495485 4.224400 4.201775 4.206081 4.272772 4.224648
#>  [81] 4.285144 4.281821 4.197711 4.106202 3.772960 4.438107 4.109416 4.223305
#>  [89] 4.276146 4.098306 4.374069 4.267839 4.319815 4.166670 4.748231 3.899360
#>  [97] 4.014729 4.285117 4.251223 4.097764
#> 
#> $time
#>       train        test 
#> 0.086755037 0.002198935

Citation

@misc{https://doi.org/10.48550/arxiv.2206.09991,
  doi = {10.48550/ARXIV.2206.09991},
  url = {https://arxiv.org/abs/2206.09991},
  author = {Silva, Aníbal and Ribeiro, Rita P. and Moniz, Nuno},
  title = {Model Optimization in Imbalanced Regression},
  publisher = {arXiv},
  year = {2022},
  copyright = {Creative Commons Attribution 4.0 International}
}

modeloptimizationir's Issues

questions to use the package

Hi,
I am very interested in this package, and try it in my dataset.

Here is the output graph and the output shows the dataset is imbalanced at low and high end, is it right?

Then I try it with xgboost to predict, but the predict result seems put much importance at low end

would you please help to suggest how can improve

Thank you!
Kun