RL - Power Environment

Agent needs to collect resources and build processor where the processor will produce items. Collecting this items gives more reward to the agent.

Check previous master commits.

Run main.py in src/RL. Code your own agent and use the algorithm from the algorithm package to train the agent. Set the train to False while testing the agent.

s = Simulator(
        agent,env,trainer,
        nactions=6,
        log_message="Testing this high performance model", 
        visual_activations= True )
    print(s.visual_activations)
    s.run(1000,5000,train=False,render_once=10,saveonce=7)

Training results are stored in the logs folder. The name of the plots are given based on the timestamp the program is run. Make sure you give a appropriate log name for each training session, so you can later identify the model using the timestamp. This log messages are stored in the src/RL/logs/log.json file.

RL -Multi Agent Training Using Gatherer Environment

Latest master commit. Use the class algorithm.reinforce.MultiEnvironmnetSimulator.

#MULTI ENVIRONMENT TESTING
boxsize = 10
na = 3
n_envs = 16 
environments = [GathererState(gr=10,gc=10,vis=5,nagents=na,boxsize=boxsize,spawn_limit=5) for i in range(n_envs)]

model = StateRAgent(input_size=100,state_size=3,containers=len(environments))
model1 = StateRAgent(input_size=100,state_size=3,containers=len(environments))
model2 = StateRAgent(input_size=100,state_size=3,containers=len(environments))

models = [model,model1,model2]
s = MultiEnvironmentSimulator(
    models,environments,nactions=6,
    log_message="Testing with 4 Environments",
    visual_activations=True)

train = True 
s.run(1000,500,train=train,render_once=1,saveonce=2)