search

notadamking / RLTrader

A cryptocurrency trading environment using deep reinforcement learning and OpenAI's gym
https://discord.gg/ZZ7BGWh
GNU General Public License v3.0
1.71k stars 537 forks source link

Any pointers of how this would adapt to live data? #67

Closed sph3rex closed 5 years ago

sph3rex commented 5 years ago

Even though is a great intro to RL stuff and pretty well documented by the articles, I can't get my head around how live data would fit the picture here, so I'm guessing that _next_observation needs some tweaking/changing, but I'm not really sure of how to tackle the whole thing, because probably I'm missing some basics here :). Any pointers would be appreciated (maybe some examples from openai's gym repo that I've missed). Thank you.

nachovoss commented 5 years ago

I think one approach would be, when running create 2 dataframes one with the preloded coinbase_hourly, and a new one with the new data , then try dataFrame.append(other, ignore_index=False, verify_integrity=False, sort=TRUE), and work on a loop constantly updating the resulting df , making prediction, act on prediction, then remember to erase the last row in the df as we just added one and erase unused variables every iteration of the loop to avoid running out of ram. Haven't tested yet it's just an idea, hope it helps

notadamking commented 5 years ago

Closing this in favor of keeping the discussion in #21.