L110_SupportVectorMachines_Template.Rmd

---
title: "Support Vector Machines"
author: "Bert Gollnick"
output: 
  html_document:
    toc: true
    toc_float: true
    number_sections: true
    code_folding: hide
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = T, warning = F, message = F)
```

```{r}
library(dplyr)
library(ggplot2)
library(keras)
library(e1071)
source("./functions/train_val_test.R")
```

# Data Preparation

## Data Creation

We create data on our own. 

```{r}
set.seed(123)
df <- dplyr::tibble(x = runif(n = 2000, min = -10, max = 10),
                              y = runif(n = 2000, min = -10, max = 10)) %>%
  dplyr::mutate(z = x^2+y^2) %>% 
      dplyr::mutate (class = ifelse(z < 50, "A", "B")) %>% 
      dplyr::mutate(class = as.factor(class))

```

You can see a circle inside a rectangular grid. The circle points refer to class A, the outer area refers to class B. The task is to predict the classes correctly.

## Train / Validation Split

Data will be splitted for 80 % training, and 20 % validation.

```{r}
c(train, val, test) %<-% train_val_test_split(df = df, train_ratio = 0.8, val_ratio = 0.2, test_ratio = 0)
```

## Visualisation

The data is visualised. It has only two dimensions, but since SVM is able to reshape the data into some higher-order representation it will be able to predict the classes nearly perfectly.

```{r}
g <- ggplot(train, aes(x, y, col = class))
g <- g + geom_point()
g <- g + theme_bw()
g
```

# Model

We create a Support Vector Machines model.

```{r}
# code here
```

It can also be plotted with the base-plotting environment.

```{r}
# code here
```

# Predictions

The classes of validation data are predicted.

```{r}
# code here
```

# Model Performance

## Confusion Matrix

A confusion matrix is created based on training and validation data.

Training:

```{r}
# code here
```



Validation:

```{r}
# code here
```

```{r}
# code here
```

## Hyperparameter Tuning

You can modify 

- kernel

- cost

- gamma

```{r}
# code here



# create predictions
val$class_pred <- predict(object = model_svm, 
                          newdata = val)

# Confusion Matrix
conf_mat <- table(predicted = val$class_pred, actual = val$class)
caret::confusionMatrix(conf_mat_val)

```