ml-remote-server
Server complement of https://github.com/getnamo/MachineLearningRemote-Unreal.
Startup the server, point your MachineLearningRemote component to your server address with the script file name set in DefaultScript property and it will load on begin play.
Quick Setup
- Install python 3 on target machine recommended version for e.g. tensorflow: 3.7.
Option 1)Using Windows Local Server
- Update requirements.txt with any dependencies you need, e.g. tensorflow==2.2
- Double click on
InstallRequirements
- (Optional) If you're not using the autolaunch option on MachineLearnineRemote: Double click
StartupServer.bat
Option 2) Using Remote Server
- Pick a folder, navigate to it
git clone https://github.com/getnamo/ml-remote-server.git- Update requirements.txt with any dependencies you need, e.g.
tensorflow==2.2 - run
pip install -r requirements.txt. - In terminal type
python server.pyto start the server
Server is now ready to use.
Optional Client Steps
- Connect your UE4 instance via https://github.com/getnamo/MachineLearningRemote-Unreal
- Listen to log events via your browser by going to
<server ip>:8080orlocalhost:8080in your browser. There are some debug commands like/r <script name>to swap script and/ito send dummy inputs; see https://github.com/getnamo/ml-remote-server/blob/master/server.py#L117 for all supported commands.
How to use
Startup event flow
Begin Play -> (if connect on beginplay) connect to backend -> (if start script on connection) Default Script start
Listen to the OnScriptStarted event to know it's safe to send inputs/start training.
API
See https://github.com/getnamo/MachineLearningRemote-Unreal as all interaction beyond debugging is handled by client.
Keep in mind that you can end play, do some code changes, and begin playing again without rebooting your server; the default script will be reloaded. You can also use the debug browser with /r <script name> or call StartScript from MachineLearningRemote component to live reload a script even during play.