pong doesn't train on a desktop

GoogleCloudPlatform / tensorflow-without-a-phd

A crash course in six episodes for software developers who want to become machine learning practitioners.

Apache License 2.0

2.78k stars 907 forks source link

pong doesn't train on a desktop #11

Open p-neq-np opened 6 years ago

p-neq-np commented 6 years ago

The pong doesn't train on my desktop with gtx 1080. The changes I made were,, the memory size was reduced to 80k from 100k and batch size is set to 128. Would this be expected?

dizcology commented 6 years ago

What were the issues you had when training on a desktop?

p-neq-np commented 6 years ago

After one day of training, the agent still played randomly and game results didn't seemed to improve (21-1). I haven't observed any memory issues but I'd thought maybe there is another set of configurations for training locally (as opposed to cloud).

martin-gorner commented 6 years ago

Were you using a GPU ?

dizcology commented 6 years ago

@brcln how many training epochs did your training job complete?