Deep Learning Your Broadband Network at Home

Explore the process of data mining and knowledge discovery for home broadband networks in this EuroPython 2017 conference talk. Learn how to automate internet speed tests, log metrics, and analyze time series data using Python. Discover techniques for finding trends, forecasting, and detecting anomalies in network performance using statistical and deep learning methods such as ARIMA and LSTM. Gain insights into handling time series data, seasonal trend decomposition, and rolling forecasts. Delve into anomaly detection approaches, from naive methods to more advanced techniques like Multivariate Gaussian Distribution. Suitable for all skill levels, this talk provides a comprehensive overview of monitoring and analyzing home network performance, encouraging Python enthusiasts to apply these concepts in their own environments.

Intro
Outline
Home Network
Anomaly Detection (Naive approach in 2015)
Problem definition
Types of anomalies in time series
Logging Data
Data preparation
Handling time series
Components of Time series data
Seasonal Trend Decomposition
Rolling Forecast
Anomaly Detection (Basic approach)
Anomaly Detection (Naive approach)
Stationary Series Criterion
Test Stationarity
Autoregression (AR)
Moving Average (MA)
Identification of ARIMA (easy case)
Identification of ARIMA (complicated)
Anomaly Detection (Parameter Estimation)
Anomaly Detection Multivariate Gaussian Distribution
Anomaly Detection (Multivariate Gaussian)
Long Short-Term Memory
Summary
Contacts
Patterns in time series