Example_GradientCOBRA.Rmd

---
title: "Example"
author: "Sothea HAS"
date: "2023-06-22"
output: html_document
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```


```{r}

if(!('pacman' %in% installed.packages()[,1])){
  install.packages('pacman')
}

pacman::p_load(MASS)
pacman::p_load(devtools)
```

## Loading the source codes with `devtools`

```{r}
devtools::source_url("https://raw.githubusercontent.com/hassothea/AggregationMethods/main/GradientCOBRARegressor.R")
```


### 1. Boston housing

```{r}
Boston |> head()

X <- Boston[c("crim", "indus", "nox", "rm", "age", "dis", "rad", "tax", "ptratio", "black", "lstat")]
y <- Boston$medv
n <- length(y)
train <- logical(length = n)
train[sample(n, floor(0.8*n))] <- TRUE
```

Using the aggregation method:

```{r}
reg <- GradientCOBRARegressor(train_design = X[train,],
                    train_response = y[train],
                    test_design = X[!train,])
```


```{r}
# Relative Root Mean Squared Error
reg$fitted_aggregate |> map(.f = ~ sqrt(mean((y[!train] - .x)^2))/mean(y[!train]))
```

You can also use more than one kernel function at a time, with other options of basic estimators and learning process as follow:


```{r}
reg1 <- GradientCOBRARegressor(train_design = X[train,],
                    train_response = y[train],
                    test_design = X[!train,],
                    kernels = c('naive', 'gaussian'),
                    optimizeMethod = c("grid", "grad"),
                    machines = c('knn', 'rf', 'xgb', 'ridge', 'lasso'),
                    setBasicMachineParam = setBasicParameter(k = 5:10,
                                                             ntree = 300,
                                                             nrounds_xgb = c(300, 500)),
                    setGradParam = setGradParameter(rate = 0.01),
                    setGridParam = setGridParameter(figure = FALSE))
```


```{r}
# Relative Root Mean Squared Error
reg1$fitted_aggregate |> map(.f = ~ sqrt(mean((y[!train] - .x)^2))/mean(y[!train]))
```


For how to you the function, see: [https://hassothea.github.io/files/CodesPhD/GradientCOBRARegressor.html](https://hassothea.github.io/files/CodesPhD/GradientCOBRARegressor.html).