danielpuka/time_series_anomaly_detection_examples

Several examples of anomaly detection algorithms for time series data.

time_series_anomaly_detection_examples

Table of Contents

• Introduction
• Python
• R
• InfluxDB

Introduction

The Time Series Anomaly Detection repo contains several examples of anomaly detection algorithms for use with time series data sets.

Examples can be found in the python directory and r directory.

InfluxDB and Grafana are optionally included in the Docker stack for data storage and visualization purposes. Telegraf has been included to fill the InfluxDB with dummy metric data.

Redis is optionally included in the Docker stack for caching (i.e. memoization) purposes.

If you're using Docker, execute build.sh to get started.

Python

Examples are typically written in python. From the .env.example file, you can see that scripts are written in python 3.8.2. A list of module dependencies can be found in the Dockerfile and requirements.txt. You aren't forced to use Docker, and can use something like Conda instead if that's your preference.

If you opt to use Docker, you can view the Makefile for relevant Docker commands. The make penter command will create a new container and execute the python CLI. The make prun command will run a python script. For example, make prun d=ts_price_anomaly_detection s=view will run ts_price_anomaly_detection/view.py

Example anomaly detection algorithms can be found in the python directory, and each example directory contains a similar structure. When exploring an example, you should first read the README.md and references.md files. The references.md file will provide you with a relevant link to a tutorial page and data set. Download the recommended data set and place it in the local data directory (don't place it in the root data directory).

You can then execute various python scripts to analyze and model the data. It's recommended that you run explore.py then view.py first to better understand the distribution of the data.

R

Additional examples are written in R. From the .env.example file, you can see that R scripts are written in version 3.6.3. A list of additional R packages can be found in the Dockerfile.

As the docker-compose.yml file shows, this repo employs the rocker/tidyverse image which already includes the tidyverse collection and RStudio server.

If you opt to use Docker, you can view the Makefile for relevant Docker commands. The make renter command will allow users to execute shell commands within the R container. The make rrun command will run an R script. For example, make rrun s=bitcoin_anomalies will run $R_STUDIO_USER/bitcoin_anomalies.r

Example anomaly detection algorithms can be found in the r directory. You can then execute various r scripts to analyze and model the data.

InfluxDB

InfluxDB is a time series database. For those who are unfamiliar, more information can be found at influxdata.com. InfluxDB can be combined with Grafana to analyze and visualize the data. View the .env.example file to configure your InfluxDB & Grafana versions and ports.

CSV files can be easily imported to your InfluxDB instance using the csv-to-influxdb package. Each example directory will contain a README.md file with a csv-to-influxdb.py command to execute.