Training Archimedes to become the most efficient yield farmer for our users.
With any ML model and when training our own, each set of input data needs to be defined and ultimately determined first. When ascertaining our ‘time series data set design,’ our team determined a length of 15 minutes for yield APY changes, while 1 minute for price data feeds.
This has specific reasoning:
- Each chain has a different update frequency;
- According to Stargate usage history (pool events analysis), 15 minutes intervals is granular enough as very few transactions occur within this time frame;
- A 1 minute feed for price data is vital in such a volatile market.