- August 14, 2020 at 10:36 am #9612adminKeymaster
Over the last few weeks, the Crypto-ML team has been hard at work on a variety of enhancements to our prediction and execution capabilities. This post will highlight a few of the details.
Our models now consider 7 new market features when building patterns and predicting outcomes. These features are descriptors of market behavior and all demonstrate the ability to be leading indicators in a variety of price change scenarios.
In order to maintain efficiency, we seek to have as few features as possible without sacrificing effectiveness. Our team tested many other features, but shortlisted to this new group of seven.
It is important to note that these additional features do not change the core behavior of our neural networks. Rather, they reinforce certain decisions and also provide better insight into a broader variety of scenarios and edge cases.
Our neural networks are now also considering additional timeframes. While trends may develop in smaller scales, it’s important to consider where those sit relative to bigger-pictures patterns as well.
This change layers in the ability to consider long-term patterns before making a prediction, ultimately helping to avoid traps and better identify breakouts.
Faster End-to-End Execution
Last, the time needed to process our entire pipeline from data ingestion (input) to prediction and trade execution (output) has been reduced by 78%. This is designed to minimize slippage and improve execution. During fast moves, every second counts.
We also continue to monitor the markets outside of our ML pipeline to identify when out-of-cycle action needs to take place.
The final versions of these updates will roll out over the next 24 hours for each of our models.
This recent bull run identified several areas of opportunity within our system. With these enhancements, Crypto-ML becomes more robust and better able to handle these exceptional situations.
Please let us know if you have any questions.
- You must be logged in to reply to this topic.