Exploring Multi-Processor Reinforcement Learning for AI Development

[@nataliescovermusicswitzerland]

😎😎😎

[@profx2020]

Do you merge patallel episode training into one model before starting the next epoch? If so, how?

[@aipricepatterns9707]

This is a really hard problem. But here are some ideas. 1. First, try using "RAY" for distributed learning. Google it There you will need to create a common object between the threads. This is very specific and important, to choose what exactly can be parallelized. If you choose a neural network and its update. And there is a question of how to update a neural network in parallel, how to understand which weights are more important than others? This is not a task for one day and maybe not even one quarter. 2. But we have thousands of different episodes and in these episodes we have a market state, an action and a result. You can collect a million different experiences from different episodes in parallel and then train a neural network based on this information. Anyway, the neural network generalizes all experience to some result and then you can follow the path of experience priority. In general, I don’t know whether you understood this or not, but it’s complicated :)

[@jaipalv562]

Hi ..do you trade options using reinforcement ..can you please make video on the same?

[@aipricepatterns9707]

Hello. I can do this if find data on options. I need historical data. For example, each episode that can be fed to the agent can be data on the expiration of options within a month. Such data batches can be fed to the agent

[@jaipalv562]

@@aipricepatterns9707 : can I give you dataset location ?

[@Madmaxlive23]

can we make a model with our trading-view indicators combinations

[@aipricepatterns9707]

Tell me in more detail what you would like to receive.