Usage

run-agent {debug,train,infer,random} [options]

Modes

Mode	Description
train	Train an RL agent against the ns-3 environment
random	Run a random agent (useful for sanity checks)
infer	Run inference using a trained checkpoint
debug	Debug mode

Options

Option	Short	Description
--env-name	-n	Name of the ns-3 environment to start
--max-episode-steps	-s	Number of environment steps per episode
--iterations	-i	Number of training iterations
--single	-sg	Use single-agent mode instead of multi-agent
--training-params	-p	key=value pairs overriding training parameters
--checkpoint-path	-a	Path to a checkpoint to load before training
--ns3-settings	-c	key=value pairs passed as CLI args to ns-3
--trainable	-t	RLlib algorithm to use (e.g. PPO)
--rollout-fragment-length	-rfl	Rollout fragment length for training
--enable-wandb		Enable wandb logging (reads WANDB_API_KEY, WANDB_PROJECT_NAME from env)
--wandb-project	-wp	Enable wandb logging to this project
--wandb-key	-wk	Enable wandb logging with this API key

Examples

Train for 100 iterations:

run-agent train --single -n defiance-lte-ns3-rl-thesis --iterations 100

Quick sanity check with a random agent:

run-agent random --single -n defiance-lte-learning --iterations 2

Train with custom ns-3 settings and wandb logging:

run-agent train --single -n defiance-lte-ns3-rl-thesis \
    --iterations 100 \
    --ns3-settings topology=multi_bs speed=10 \
    --trainable PPO \
    --enable-wandb

Resume training from a checkpoint:

run-agent train --single -n defiance-lte-ns3-rl-thesis \
    --iterations 100 \
    --checkpoint-path /path/to/checkpoint