PoC application that runs analyzis on Coinbase cryptocurrency exchange data stream. The application is created as a part of Kahoot! interview process.
The application connects to the level2_batch data feed for the given product id (currencies pair sold on the platform, for example "BTC-USD" or "ADA-EUR") and maintains an order book - a list of all sell requests (asks) and buy requests (bids) for a given exchange auction. Every 5 seconds the user sees the following book statistics:
- Highest bid price and quantity
- Lowest ask price and quantity
- What is the biggest difference between the highest bid price and lowest ask price so far
- Average mid-prices (average price of the current highest bid price and the current lowest ask price) for the last 1, 5 and 15 minutes
- Forecasted mid-price in 60 seconds
- The forecasting (absolute difference between the predicted price and the observed price once it arrives) for the last 1, 5 and 15 minutes
Note that forecasting stats aren't available from the beginning. This is because the model needs data to train on. The model is trained after the first 2 minutes of data is received. Then, it is updated every 6 seconds and completely retrained every 2 minutes. Additionally, if source data is fits the model poorly, the app will increase (re)training interval by 1 minute until the model is trained successfully. The app stores at most 15 minutes of data, so data features with longer seasonality won't be reflected.
The application can be started in the provided docker container. Run
docker build . -t coinbase-exchange-analyzer:latest
docker run coinbase-exchange-analyzer:latest [args]The application connects to the BTC-USD on wss://ws-feed.pro.coinbase.com by default. These parameters can be
changed by dedicated command line arguments. Run docker run coinbase-exchange-analyzer:latest --help for more details.
The application was developed and tested for Python 3.11.4. All required libraries are specified in requirements.txt