Skip to content

Latest commit

 

History

History
32 lines (27 loc) · 1.02 KB

docker.md

File metadata and controls

32 lines (27 loc) · 1.02 KB

Docker Setup

Overview

For training, we have set up a docker containiner that can be deployed to a server. Unity cannot be easily ran remotly because it requires a license, but a unity execuitable can be built and ran in headless mode for training the rl model.

1. Docker Hub

If you just want to pull the image from Docker Hub:

docker pull grantgasser/autonomous-vehicles-mlagents-unity:submission

2. Build and Run without Docker Hub

  • Dockerfile
    • Based on the python docker image base
    • Installs require python libraries and copy latest RL and CNN codebase
  • ./docker_build.sh x.x
    • params:
      • version (e.g. 0.1)
    • builds the docker image
  • ./docker_run.sh x.x
    • params:
      • version (e.g. 0.1)
    • run the docker image (bash)
    • NOTE: This will not grant access to the gpu. Need CUDA support

How to train

RL

  • run the docker image
  • cd rl-training
  • ./train.sh $0 $1
    • $0 = runid (e.g. test01)
    • $1 = number of environment (e.g. 1)
      • Only run more environments if using the GPU (we trainined with 6)