L130_Dbscan_Template.Rmd

---
title: "Dbscan"
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 = TRUE, warning = F)
```

# Data Preparation

We will create a dataset, that consists of a circle in the center and a ring that surrounds the inner circle.

```{r}
suppressPackageStartupMessages(library(dplyr))
suppressPackageStartupMessages(library(ggplot2))
suppressPackageStartupMessages(library(dbscan))
```


```{r}
set.seed(123)
n_points <- 4000

df <- tibble(x = runif(n = n_points, min = -10, max = 10),
             y = runif(n = n_points, min = -10, max = 10),
             z = x^2+y^2) %>% 
  mutate(class = ifelse(z < 10, "A", "B")) %>% 
  mutate(class = as.factor(class)) %>% 
  filter((z <10) | (z > 80 & z < 100))
```

Let's visualise the data.

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

# Modeling: dbscan

We start the modeling.

At first we transform the dataframe to a matrix.

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

We need to get an estimate of eps parameter.

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

For epsilon you choose the maximum value which has a low slope. 

It still requires some manual adaptation to find the optimum set of parameters.

```{r}
# code comes here

g <- ggplot(df, aes(x, y, col = factor(cluster)))
g <- g + geom_point()
g
```

# Modeling: kmeans

Just for comparison we perform kmeans to see how this performs on this dataset.

```{r}
# code comes here

g <- ggplot(df, aes(x, y, col = factor(cluster_kmeans)))
g <- g + geom_point()
g
```