---
title: "How to use breakDown package for SVM models"
author: "Przemyslaw Biecek"
date: "`r Sys.Date()`"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{model agnostic breakDown plots for SVM model}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r setup, include = FALSE}
knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)
```

This example demonstrates how to use the `breakDown` package for models created with the [kernlab](https://CRAN.R-project.org/package=kernlab) package. 


```{r}
library("breakDown")
library(kernlab)

wine_svm_model <- ksvm(quality~., data = wine)
wine_svm_model

# or with the e1071:::svm

library(e1071)
wine_svm_model <- svm(quality~., data = wine)
wine_svm_model

```

Now we are ready to call the `broken()` function.
Since `kernlab` is useing S4 methods we need to pass here the hook to `kernlab:::predict` method.

```{r}
library("breakDown")
nobs <- wine[5, , drop = FALSE]
base_prediction <- predict(wine_svm_model, nobs)
set.seed(1313)

explain_5_up <- broken(wine_svm_model, new_observation = nobs, 
                    data = wine, predict.function = predict,
                    baseline = "intercept", direction = "up")
explain_5_up

explain_5_down <- broken(wine_svm_model, new_observation = nobs, 
                    data = wine, predict.function = predict,
                    baseline = "intercept", direction = "down")
explain_5_down
```

And plot it.

```{r, fig.width=7}
library(ggplot2)
plot(explain_5_up) + ggtitle(paste0("Prediction for SVM model ", round(base_prediction, 3)))
plot(explain_5_down) +  ggtitle(paste0("Prediction for SVM model ", round(base_prediction, 3)))
```