ReMu-Agent-Env 3000

Developing an Environment for Neural Network based Reinforcement Multi-Agent Interaction

Server Usage:

python server.py [-h] [--addr | -a ADRESS] [--port | -p PORT] [--verbose | -v]

example usage:

python server.py --addr "192.168.2.420" --port 1337 --verbose

Server Arguments

Argument	Type	Description
-h, --help	None	shows argument help message
-a, --addr	STR	specifies the address of the server (default=localhost)
-p, --port	INT	specifies the port of the server (default=1337)
--training_mode	BOOL	sets the server to training mode which updates once all clients provide their action
-v, --verbose	BOOL	flag to set the server to verbose mode

Client Usage:

python client.py [-h] [--name | -n NAME] [--addr | -a ADRESS] [--port | -p PORT] [--verbose | -v] |--spectate]

example usage:

python client.py --name "Dieter" --addr "192.168.2.420" --port 1337 --verbose

Client Arguments

Argument	Type	Description
-h, --help	None	shows argument help message
-a, --addr	STR	specifies the address of the server on which the client tries to connect(default=localhost)
-p, --port	INT	specifies the port of the server on which the client tries to connect (default=1337)
-v, --verbose	BOOL	flag to set the client to verbose (logging) mode
-n, --name	STR	specifies the name of the player (it's ID)
--spectate	BOOL	starts the client in spectator mode without a starship

Neural Network Usage:

python network.py [-h] [--name | -n NAME] [--n_models | -nm N_MODELS] [--addr | -a ADRESS] [--port | -p PORT] [--verbose | -v] [--param_search | -ps]

example usage:

python network.py --name "model_0" --addr "192.168.2.420" --port 1337 --device "cuda:1" --verbose

Network Arguments

Argument	Type	Description
-h, --help	None	shows argument help message
-a, --addr	STR	specifies the address of the server on which the model tries to connect(default=localhost)
-p, --port	INT	specifies the port of the server on which the model tries to connect (default=1337)
-v, --verbose	BOOL	flag to set the model to verbose (logging) mode
-n, --name	STR	specifies the name of the model.
-m, --model_type	STR	type of the model (e.g. "linear" or "lstm")
--test	BOOL	Sets network to testing modus
-d, --device	STR	specifies the device on which the model should be trained (e.g. "cpu" or "cuda:x", default="cuda:0"). Can be used to also specify the specific GPU (e.g. cuda:2)
-ps, --param_search	BOOL	flag to activate parameter search mode (which will use the parameter dictionaries from the constants.py file)

Spawn multiple networks simultaneously via bash-script:

where --num_models flag defines the number of models to spawn. The networks will train on all available (loaded) nvidia GPUs.

chmod +x spawn_networks.sh
./spawn_nets.sh --num_models [N] --model_type [linear | lstm | cnn] --addr [IP] --device [cpu | cuda:x]