Saturday 21 Sep, 2019

Starting at 9am

Etc Venues

One Drummond Gate, Victoria, London SW1V 2QQ
Menu

Anomaly Detection Training – Peter Tillotson

Session Description

With my Security hat on, and with data streaming through a real time pipeline I wanted to know when I was receiving unusually rates of messages from given sources. But what was unusual and how do I accommodate for peoples weekly working patterns.

Enter Twitter’s Seasonal ESD algorithm ( https://github.com/twitter/AnomalyDetection) but that was written in R, I needed a version in Python. With coffee and some great Panda’s and Python libraries it turned out to be pretty straight forward. 

Learning objectives    

In the 3 hour session we will:

  • Use NAB anomaly data sets (https://github.com/twitter/AnomalyDetection)
  • Load the CSV data into Python Pandas
  • Learn to use interpolation to fill holes in data
  • Explore Linear Regression and windowed stats approaches
  • Finally we’ll use Seasonal ESD

About the trainers

Peter Tillotson is a Data Scientist at Riot Games and have worked there for 6 years in both Security and Insights groups. I love hard problems like understanding human behaviour or by trying to get my algorithms running fast enough on large datasets so that I can spend more time innovating and less time waiting while machines heat up. 

Session Requirements

Skills:

  • Some Python programming
  • Knowledge of Panda’s will definitely help but not critical
  • Analytics / Stats / Numerate background will also help

Hardware: 

  • Laptop (It is going to be a practical session you’ll be doing most of the work)  
  • Python ( >3.6, we need to let go of 2.7 )
  • We will probably be working with Jupyter notebooks so creating a clean virtual environment with `jupyter` and `pandas` installed is going to be step 1 of the session.