Published on

Time Series Modelling

Authors

I was obsessed with the idea of trying to forecast the price of etherium. The following is my disocvery of how different approaches to applying time series analysis to etherium modelling. The following is a summary of a talk I provided to the Melbourne big data meetup and Python meetup at ANU in Canberra. Please see my slides below:

[https://slides.com/andrewcarr-1/tsmodelling/embed]

The first approach I took is using facebook/meta’s package called Prophet. It is the most basic forecasting tool which only takes time as an identifier and a dependent variable, like Ethyreum price. The package has great documentation and was easy for me to get started. I also liked the visualisations that it produced which demonstrated different seasonal, long term and changing trends.

The second approach I used used freqtrade. I originally found freqtrade through Part TIme Larry (below). I found freqtrade asy to setup and get started using docker-compose. It seemed to work by starting a docker container for each strategy. The tool uses a CLI as the means of testing strategies, backtesting and deploying strategies. It also interacts well with binance. Where I struggled was getting the prophet package to load within the docker image. As a result of struggling to import machine learning libraries I lost interest.

https://www.youtube.com/watch?v=H2OkrvSojOI

For future time series modelling use cases I want to test the TSAI package as it is a FastAI libarary with a similar training inference model as the rest of thier tools. By memory it also provides the ability to test a model against a portion of random time segments.