---
title: "caret - Neural Network"
output: html_document
vignette: >
  %\VignetteEncoding{UTF-8}
  %\VignetteIndexEntry{caret - Neural Network}
  %\VignetteEngine{knitr::rmarkdown}
editor_options:
  chunk_output_type: console
---

```{r, include = FALSE}
knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "vig/"
)
options(rmarkdown.html_vignette.check_title = FALSE)
```

<!-- avoid border around images -->
<style>
img {
  border: 0;
  }
</style>

This guide is designed as a quick-stop reference of how to use some of the more popular machine learning R packages with `vivid`.  In the following examples, we use the air quality data for regression and the iris data for classification. 
  
```{r setup, echo = FALSE, warning=FALSE, message=FALSE}
library('vivid')
library('caret')
```



### caret - Neural Network


The `caret` package (short for Classification And REgression Training) in R provides a unified interface to streamline the process of creating predictive models. In the following example, we use the `caret` package to utilize a neural network model fit via the `nnet` package. As `caret` is catered for in `vivid`, there is no need for a custom predict function here. 

### Regression
```{r, message=FALSE, warning=FALSE, eval = F}
# load data 
aq <- na.omit(airquality)

# build caret nnet model
nn <- train(Ozone ~ ., data = aq, method = "nnet", trace = FALSE, linout = TRUE, maxit = 200)

# vivid
vi <- vivi(data = aq, fit = nn, response = 'Ozone')

```

#### Heatmap
```{r, caret_r_heat,  out.width = '100%', eval = F}
viviHeatmap(mat = vi)
```

```{r, echo = F,  out.width = '100%'}
knitr::include_graphics("https://raw.githubusercontent.com/AlanInglis/vivid/master/vignettes/vig/caret_r_heat-1.png")
```
<center>
<caption><span id="fig1:fig1">Figure 1: </span>Heatmap of a caret neural network regression fit displaying 2-way interaction strength on the off diagonal and individual variable importance on the diagonal.</caption>
  </center>


#### PDP
```{r, caret_r_pdp, out.width='100%', eval = F}
pdpPairs(data = aq, 
         fit =  nn, 
         response = "Ozone", 
         nmax = 50, 
         gridSize = 4,         
         nIce = 10)
```

```{r, echo = F,  out.width = '100%'}
knitr::include_graphics("https://raw.githubusercontent.com/AlanInglis/vivid/master/vignettes/vig/caret_r_pdp-1.png")
```
<center>
<caption><span id="fig2:fig2">Figure 2: </span>Generalized pairs partial dependence plot for a caret neural network regression fit.</caption>
</center>

### Classification

```{r, eval = F}
# Load the necessary library
library("caret")


# Load the iris dataset
data(iris)
irisMod <- iris
# Convert the Species to a binary classification: Setosa or not Setosa
irisMod$Species <- as.factor(ifelse(irisMod$Species == "setosa", "setosa", "not_setosa"))


# Train a neural network model
nn <- train(Species ~ ., data = irisMod, method = "nnet", 
               trControl = trainControl(method = "cv", number = 5), 
               tuneLength = 1,
               linout = FALSE, # this is set to FALSE for classification problems
               trace = FALSE, 
               maxit = 200)


vi <- vivi(data = irisMod, fit = nn, response = 'Species')

```


#### Heatmap
```{r, caret_c_heat, out.width='100%', eval = F}
viviHeatmap(mat = vi)
```
  
```{r, echo = F,  out.width = '100%'}
knitr::include_graphics("https://raw.githubusercontent.com/AlanInglis/vivid/master/vignettes/vig/caret_c_heat-1.png")
```
<center>
<caption><span id="fig3:fig3">Figure 3: </span>Heatmap of a caret neural network classification fit displaying 2-way interaction strength on the off diagonal and individual variable importance on the diagonal.</caption>
  </center>
  
#### PDP  
```{r, caret_c_pdp, out.width='100%', eval = F}
pdpPairs(data = irisMod, 
         fit =  nn, 
         response = "Species", 
         nmax = 500, 
         gridSize = 20,         
         nIce = 100,
         class = 'setosa')

```

```{r, echo = F,  out.width = '100%'}
knitr::include_graphics("https://raw.githubusercontent.com/AlanInglis/vivid/master/vignettes/vig/caret_c_pdp-1.png")
```
<center>
<caption><span id="fig4:fig4">Figure 4: </span>Generalized pairs partial dependence plot for a caret neural network classification fit.</caption>
</center>