SUPERALGOS GOVERNANCE

Luis’s Proof of Value, March 2022

Adding a new Use Case for the Superalgos SA Token.

Since March 2022 I decided to support the project via creating use cases for the Superalgos SA token. Superalgos is already a large system, all free and open source. All use cases of the token are living in the future, subject to great user adoption. What if we could have a short term use case for the token?

Machine Learning

A few years ago I tried to integrate Tensor Flow into Superalgos. The javascript version of TensorFlow was not ready by then, and the task was never finished. This March I dived into the native version for Python and start studying how to use it on strategies. Soon I realized that it is not enough to be able to use machine learning technology. For a ML model to be performant, you need to train it with the right parameters and data. Training takes a lot of processing power and time. And… how can you know which are the right parameters and data to produce the best possible performant ML model?

Nobody knows the answer to that question. So even if you took the time to learn Python, Tensor Flow, and many other related technologies to be able to train a ML model by yourself, you will hit the wall once you realize that your model’s % error is too high to be usable in an automated strategy. Nothing prevents you to try to train your model yourself with different parameters and different data, except that it will take you a lot of time, processing power, and if you do it at the cloud, a lot of money too.

Opportunity

Every problem is in the end an opportunity for whoever finds a solution to it, and I believe I did. The solution is a system that automates the testing of all possible parameter and data sources, in the quest to find the right combination of them for each crypto asset, at a certain timeframe. During March 2022, that is exactly what I did. Now I’ve got the system and it is working.

I built it in a way that instead of running it all by myself with the small set of computers I have at home, I can allow anyone to run one of it pieces, the Test Server, so that we can distribute the task of testing each combination of parameters and data between anyone who wants to participate.

Core Idea

The idea is simple, a crowd test each parameters /data combination using their own hardware, and after each test is complete, each one receives a list with the best forecasts at the moment, calculated with the best performing ML models discovered at that time by the crowd that is testing. These forecasts are saved in Superalgos as a new indicator that can be used in strategies.

Once a set of parameters and data is discovered to have better performance than the previous one known, then the last replaces the previous, and all the forecasts for that asset / timeframe are then calculated with the best performant ML model from there on, until it is eventually replaced again by some other ML model in the future.

As you can see, the accuracy of predictions should improve overtime.

The Code

The code of the system lives mostly outside the Superalgos project, and belongs to the Bitcoin Factory project, a spin off Superalgos and the first entity of the Superalgos ecosystem. There is some code inside Superalgos, that helps with the integration between these two systems.

• Bitcoin Factory Server (inside the Platform Client)
• Bitcoin Factory Data Mine (with the indicator and plotter that will allow you to use the forecasted candles in your strategies)

The Token Use Case

Being able to know which is the min, max and close of the next candle is a big deal, and something that anyone would like to know. Of course it will take time to lower the ML models % error, but as I explained, they can only improve over time.

What if we required a certain token power to get the predictions? Well, that is exactly what we are going to do. We are starting with no requirements at version 1, but as time goes by, the system gets improved and the predictions become more accurate, getting access to the stream of predictions will require token power. How much exactly we don’t know yet, whatever makes sense. We can discuss this later down the line.

Conclusion

This opens a new era for the Superalgos project, were other entities can become part of it’s ecosystem adding more use cases for the token. I personally have a few more ideas that will explore in the future. In the meantime, I will refine this system to get the best possible ML models that produces the most accurate candles forecasts for the crypto world.