README.md

In berdaniera/markr: markr / Human-assisted anomaly detection

markr

Marking time series data in R

markr provides a visual interface for rapid anomaly detection in time series data. It can be used as part of an analysis work flow or stand alone.

The package fits a machine learning algorithm (support vector machine) to automatically suggest anomalies. The user can keep those flags or mark new sections of the data and add flag names and comments. Marked data and flag metadata can be saved for future use. The fitted model objects can be saved locally for training future data sets as well.

Installation

Add a magic marker to your work flow

The package can be installed with this R command:

devtools::install_github("berdaniera/markr")

Usage example

markr can easily be integrated into your existing analysis:

library(markr)
# set working directory for output
setwd(".")
# load example data
data(markeg)
# match required data format for marking
data = tidyr::spread(markeg, variable, value)
# mark it up
mark(data)

The interface can also be used without any pre-loaded data. You can open the UI and upload a new file by calling:

mark()

Input data

The data that you enter must meet the formatting requirements:

• a first column with a date-time format named 'DateTime'
• additional columns with variables
• one row for each date-time
• only one header row

Outputs

Flagged data can be stored for future use in three ways:

1. Saving flagged data for training future datasets from the same data source (in a subdirectory of your current working directory).
2. Saving a copy of your original dataset with a new 'Flag' column and flag metadata as .csv files in your current working directory.
3. Returning a list (markOut) to your current R workspace that includes the flagged data (data), flag metadata (flags), and the model object (model).

Dependencies

This package requires the following additional packages to work correctly: dplyr, e1071, ggplot2, readr, shiny, tidyr.

FYI

Aaron Berdanier – @berdaniera – [e-mail]

Distributed under the GNU General Public license. See LICENSE for more information.

https://github.com/berdaniera

berdaniera/markr documentation built on May 12, 2019, 3:04 p.m.