Performance issue about deep-trading-agent HOT 6 CLOSED

what should I configure to let your code to run on GPU for tensorflow?

from deep-trading-agent.

Are you running the script on a hardware accelerator (eg. GPU or TPU) with tensorflow configured for its use?

from deep-trading-agent.

Not yet, I only try with the CPU version. Anyway the it/s starts at good rate almost 3000 it/s but then, after 10 minute collapse at 20 it/s. Is his normal or happens only to me?

By the way you can mark this as question ;)

from deep-trading-agent.

Yeah it initially starts with a high rate because it takes random actions upto certain number time steps after which it starts to back-propagate the losses through the network. Thus, the sudden drop in the number of iterations.
Further, it is slow when run only on CPU and won't have significant speed up on GPU as well since the computation is stalled every time the next action is evaluated.

from deep-trading-agent.

Then i need to wait 10 days of computation ?!?

from deep-trading-agent.

There is still work left to tune the hyper parameters of the model and to decide upon the most suitable reward function. Currently the model converges after around 2-3 hours to a stable reward value (though pretty close to 0) .

from deep-trading-agent.