README.Rmd

---
output: github_document
---

<!-- README.md is generated from README.Rmd. Please edit that file -->

```{r, include = FALSE}
knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "man/figures/README-",
  out.width = "100%",
  message = FALSE,
  warning = FALSE
)
```

# anomalyrecipes

<!-- badges: start -->
<!-- badges: end -->

Anomaly recipes provides ```{recipes}``` step functions for anomaly detection. 

Right now the package contains one step function: ```step_isofor``` (short for isolation forest). This is an implementation of the isolation forest algorithm from the R ```{solitude}``` package. The plan is to also add a matrix profile step function from the ```{tsmp}``` package.

This is a work in progress (and mostly a pet project at this point), but if you have a chance to try it out, please let me know how it goes. 

## Installation

You can install the development version of anomalyrecipes from github:

``` r
devtools::install_github("kevin-m-kent/anomalyrecipes")
```

## Getting Started

Let's start with a case where we are building some pre-processing steps for a linear model. We would like to add a feature that captures how anomalous observations are. 

```{r step_isofor_example, message=FALSE, warning=FALSE}
library(anomalyrecipes)
library(tidymodels)
library(tidyverse)
library(solitude)

tidymodels_prefer()

splits <- initial_split(mtcars)
train <- training(splits)
test <- testing(splits)
resamples <- bootstraps(train, times = 5)

rec_obj <- 
  recipe(mpg ~ ., data = mtcars) %>%
  step_dummy(all_nominal_predictors()) %>%
  step_isofor(all_predictors(), sample_size = 10, max_depth = 5)

baked_data <- rec_obj %>%
  prep() %>%
  bake(train)

baked_data %>%
  head()

```
You can also tune the isolation forest pre-processing step:

```{r tuning,  message=FALSE, warning=FALSE}
lm_mod <- linear_reg() %>%
  set_engine("lm") 

rec_obj_tuned <- 
  recipe(mpg ~ ., data = mtcars) %>%
  step_dummy(all_nominal_predictors()) %>%
  step_isofor(all_predictors(), sample_size = tune(), max_depth = tune())

wf_linear <- workflow() %>%
  add_recipe(rec_obj_tuned) %>%
  add_model(lm_mod)

iso_param <- wf_linear %>%
  parameters() %>%
  update(sample_size = anomalyrecipes::sample_size(c(1, 24)))

tuned_results <- wf_linear %>%
  tune_grid(resamples = resamples, param_info = iso_param)

tuned_results %>%
  autoplot()

```