Repository files navigation

• Ubuntu 16.04
• tmux
• TensorFlow >=1.0.0
• Basic knowledge on tmux is required, e.g., shortcut keys.

• For users with single GPU, do the following command:

$ bash run_single_gpu.sh

Inrun_single_gpu.sh, you can increase the number of workers by modifyingnum_workers.

• For users with multiple GPUs, do the following command:

$ bash run_multi_gpu.sh

Inrun_multi_gpu.sh, you can increase the number of workers by modifyingnum_workers.Note that the size ofGPU_ID inrun_multi_gpu.sh should be the same asnum_workers.For example, ifnum_workers is equal to 2,GPU_ID might be(0 1),(2 4), ...

• The network architecture from@ischlag. However, not exactly the same. For example, TensorFlow graph is a little bit different.
• Some functions and ideas come from OpenAI'suniverse-starter-agent. However, the original code does not support GPU usage.