Mirocana Development Update

Mirocana Development Update

George (Mirocana CEO)

Dear Token Holders!

We, at Mirocana.com, are trying to meet deadlines that were set in our roadmap. The process of development of sophisticated predictive software is tricky by its design. We underestimated the amount of work that is needed to be done before we are ready to let the system to manage real assets of our token-holders. In addition to previous Mirocana Development Update, I would like to share some insights on the Mirocana structure and internal logic.

We have spent the last month to redesign the system to more stable, resilient and scalable version of itself. Among other things we have implemented several features that are essential for a successful long-term operation: 

  • The historical data of predictions and super-predictions is now calculated on worker-nodes which allows us to deploy the system to new servers in a matter of minutes and unlocks substantial computational power that is needed. 
Without that feature, we were unable to successfully scale the system and were limited by the certain number of strategies to be included in super-predictions. 
  • The implementation of main-node allowed us to control worker-nodes and ensure unique sets of data with no duplicates. This configuration allows us to implement multiple instances of ClickHouse (database we use) and connect them to a single main-node. 
Without that feature, we were unable to control the entire system from a singular standpoint and ensure safe restore in case of a hard crash of the system which could unpredictably affect the consistency of data. 
  • The logic of super-strategies was implemented. Before that, we were training neural nets and predictive models on a simulations layer of the system. 
Without that feature, we were unable to test and evaluate the similar logic of trades opening with a different set of parameters because each run would take up to 6 hours. Now we are able to test many logics in a matter of minutes. 

Old / New Logic Comparison

What was the problem: It was expected that we will be able to implement all of these new features in January and conduct all required testing before February. The amount of work was underestimated and several issues have occurred on the way: 

  • The control of worker-node required a creation of management system (admin panel) for all docker containers that we are running. Without the ability to stop, restart and redeploy all or some of the containers with certain settings at push of the button we were unable to easily operate and debug the system. We failed in attempt to create our own management system and switch to an open-sourced solution with some minor edits.
  • The logic of tasks prioritization and management also took longer than expected. We have implemented a new entity - task manager - that is responsible for controlling the equal distribution of tasks among all worker-nodes using the RabbitMQ queue.  
  • To implement super-strategies logic we were had to reinvent the logic of simulations. Now we have to types of simulations: 1) allocating simulations that allocate free capital among different assets or asset groups. 2) trading simulations that are using buying power to long/short various assets.

We intend to improve our ability to fairly estimate the time consumption of new tasks using that negative experience.

Future plans and release dates: It is known that we would like to make our token transferable to ERC20 wallets as soon as we release our first product: trading integration with Bittrex. Our deadlines are:

  • Publish the results of system's historical performance of management by 28th of February 2018 or earlier. We have intermediate results but still would like to improve them by then and finalize all simulations. 
  • The launch of operation of the system with best performing simulation on all connected Bittrex accounts of customers by 28th of February or earlier. We would like to take a bit of a more time to ensure no bugs while testing the system on our accounts. 
  • Release the ability to transfer MIRO tokens to ERC20 wallets by 6th of March 2018 or earlier. The interface is available at app.mirocana.com but this feature is switched off server-side. We are preparing marketing activity and articles to support that news.

I apologize in front of all token holders for the delays in the development of the system. We have a strong vision for the long-term development of the technology that is partly reflected in this document. As we start to successfully manage the funds of token-holders the strong demand for the token (access key for our products) will appear in a matter of weeks thanks to the interconnectedness of crypto-community. Our performance results will only improve as we add more data, strategies and test new configurations of neural nets.