README.Rmd

---
output: github_document
bibliography: vignettes/bibliography.bib
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE,
                      fig.path="man/figures/")
```
# dobin <img src='man/figures/logo.png' align="right" height="132.5" />

<!-- badges: start -->
  [![R-CMD-check](https://github.com/sevvandi/dobin/workflows/R-CMD-check/badge.svg)](https://github.com/sevvandi/dobin/actions)
  <!-- badges: end -->

[![](https://cranlogs.r-pkg.org/badges/dobin)](https://cran.r-project.org/package=dobin)

The R package **dobin** constructs a set of basis vectors tailored for outlier detection as described in [@dobin]. According to Collins English dictionary, "dob in" is an informal verb meaning to inform against specially to the police. Naming credits goes to Rob Hyndman.  

## Installation
You can install dobin from CRAN:
```{r ins1, eval=FALSE}
 install.packages("dobin")
```

Or you can install the development version from [GitHub](https://github.com/sevvandi/dobin)
```{r ins2, eval=FALSE}
 install.packages("devtools")
 devtools::install_github("sevvandi/dobin")
```

## Example

A bimodal distribution in six dimensions, with 5 outliers in the middle.  We consider 805 observations in six dimensions. Of these 805 observations,  800 observations are non-outliers;  400 observations are centred at $(5,0,0,0,0,0)$ and the other 400 centred at $(-5,0,0,0,0,0)$.  The outlier distribution consists of $5$ points with mean $(0,0,0,0,0,0)$ and standard deviations $0.2$ in the first dimension and are similar to other observations in other dimensions.

```{r bimodal}
library(dobin)
library(ggplot2)

set.seed(1)
# A bimodal distribution in six dimensions, with 5 outliers in the middle.
X <- data.frame(
   x1 = c(rnorm(400,mean=5), rnorm(5, mean=0, sd=0.2), rnorm(400, mean=-5)),
   x2 = rnorm(805),
   x3 = rnorm(805),
   x4 = rnorm(805),
   x5 = rnorm(805),
   x6 = rnorm(805)
)
labs <- c(rep("Norm",400), rep("Out",5), rep("Norm",400))
out <- dobin(X)
autoplot(out)

```

To see the outliers in a different colour we plot again. 
```{r plotting}
XX <- cbind.data.frame(out$coords[ ,1:2], as.factor(labs))
colnames(XX) <- c("DB1", "DB2", "labs" )
ggplot(XX, aes(DB1, DB2, color=labs)) + geom_point() + theme_bw()


```

To compare, we perform PCA on the same dataset. The first two principal components are shown in the figure below:

```{r bimodal_pca}

set.seed(1)
# A bimodal distribution in six dimensions, with 5 outliers in the middle.
X <- data.frame(
   x1 = c(rnorm(400,mean=5), rnorm(5, mean=0, sd=0.2), rnorm(400, mean=-5)),
   x2 = rnorm(805),
   x3 = rnorm(805),
   x4 = rnorm(805),
   x5 = rnorm(805),
   x6 = rnorm(805)
)
labs <- c(rep("Norm",400), rep("Out",5), rep("Norm",400))
out <- prcomp(X, scale=TRUE)
XX <- cbind.data.frame(out$x[ ,1:2], as.factor(labs))
colnames(XX) <- c("PC1", "PC2", "labs" )
ggplot(XX, aes(PC1, PC2, color=labs)) + geom_point() + theme_bw()

```

# References