SUPERALGOS GOVERNANCE
Luis Proof of Value, April 2022
Bitcoin Factory integrated into Superalgos and stabilised.
This month I completed a new iteration on the machine learning front of the project. Continuing the work of March, that you can see here:
April was the month that I integrated the Bitcoin Factory pieces of software into Superalgos, and stabilised the software so that it can run smoothly.
The previous version required to clone a separate repository, and manually set an identity and many other things.
After the integration, everything is coming inside the Superalgos codebase, and less configuration steps are required.
During April we reached the 4th round of Tests, and finally we could run all the system components: the Test Clients, Forecasters, Test Server and Superalgos Network Node without issues.
We are ready for you to join!
Take a look at the Bitcoin Factory README file for instructions on how to setup your Test Client:
There is an interesting discussion going on at the Machine Learning Telegram Group. If you are not following it, jump in and catch up.
If we all contribute some processing power we can find ML models that can produce better forecasts faster. It is time to join now! The polishing has already been done and there is a cool group of people involved and willing to help you with the setup in case you need it.
It is already working!
Check out the rounds #3 and #4 of testing and see by yourself how we managed to find a model to predict 1hour BTC/USDT candles with a 0.67% of error.
If you like analysing data, we need you! We need a team of people that can arrive to conclusions based on these reports.
The problem to solve is that there are millions of combinations of parameters and datasets to test and it is impossible to test them all. We need to reduce that test space by discarding parameters values that are not improving the results, but that is difficult to spot. We also need a team that designs each test round based on the results of previous rounds. If you would like to be part of such a team, please let me know.
Add as many processing power as you like!
As you can see at the following chart, there are user profiles running more than one instance of the Test Client at the same time.
Integration into Governance
We are looking for a dev to integrate the test reports into the Governance System so that from next month the ones that have been running Test Client instances can be compensated with SA tokens through a new Governance program for this purpose.
Bastian Dombret is that you?
Setting up all Indicators
theblockchainarborist is compiling a list of Indicator properties from the different Data Mines / Indicators we already have that could be used on the training of these ML models. If you have created indicators, please help him to pick which indicator properties could be useful for this. We need properties that are not correlated to each other and that are not intended simply to support the plotting of data or to be used only at a trading system. The sooner we get this full list, the sooner we can run more test rounds.
Forecast Signals Coming Soon!
Next on my roadmap is to broadcast these forecasts via the Superalgos P2P network Trading Signals infrastructure. That is when we will be completing the loop and having the first SA token use case, since these signals will be available only to SA token holders, with some rules according to the amount held (details needs to be though out; the only thing that is clear is that we will have the signals for ourselves, the ones that actually contributed to the project and are holding SA tokens).
Test Round #5 Coming Very Soon!
Get your Test Client setup soon and participate at the next test round, starting early next week.
Training of an Agent Via Reinforcement Learning Coming Soon!
As you can see at Alexandru’s proof of value report, the research phase for what is needed to train a profitable agent to trade crypto has been completed and the implementation is about to start. That includes the integration with the current infrastructure of the Bitcoin Factory project so that the training and best parameters discovery can be crowd sourced. We will need much more processing power for this, since the preliminary tests shows that at least 30 million iterations over the episode are needed to start seeing good results. That could be 3 or more days on a single hardware, and we need hundreds of tests for different parameters and datasets. Please get involved by providing processing power. There are already developers investing quite a lot of energy to pull this out, do as much as you can setting up all your idle processing power you have at home.
Devs With Experience in Machine Learning Needed!
If you have experience in ML, please get involved reviewing the code and what we are doing in general. That could save us time if we might be going in the wrong direction. Get to the ML Telegram Group and join the conversation.
Conclusion
Machine Learning is the hottest part of the project right now. Get involved and help move it forward.
If you are already familiar with the system, running a Test Client will be no issue for you. One instance usually consumes 30% of CPU, the same as the Platform UI, so not a big deal, specially if you have idle computers at home, you can just leave it running there, unattended.
