criticOptimizerTwo = optim.Adam(criticNetTwo.parameters(),
lr=config['criticLearningRate'])
actorNets = {'actor': actorNet, 'target': actorTargetNet}
criticNets = {
'criticOne': criticNet,
'criticTwo': criticNetTwo,
'targetOne': criticTargetNet,
'targetTwo': criticTargetNetTwo
}
optimizers = {
'actor': actorOptimizer,
'criticOne': criticOptimizer,
'criticTwo': criticOptimizerTwo
}
agent = TD3MultiStageUnit(config, actorNets, criticNets, env, optimizers,
torch.nn.MSELoss(reduction='mean'), N_A)
agents.append(agent)
controller = MultiStageStackedController(config, agents, env)
loadCheckPoint = False
if loadCheckPoint:
checkpoint = torch.load('Log/Stage1Finalepoch8000_checkpoint.pt')
controller.agents[1].actorNet.load_state_dict(
checkpoint['actorNet_state_dict'])
controller.agents[1].actorNet_target.load_state_dict(
checkpoint['actorNet_state_dict'])
controller.agents[1].criticNetOne.load_state_dict(
checkpoint['criticNetOne_state_dict'])
controller.agents[1].criticNet_targetOne.load_state_dict(
checkpoint['criticNetOne_state_dict'])
actorNets = {'actor': actorNet, 'target': actorTargetNet}
criticNets = {
'criticOne': criticNet,
'criticTwo': criticNetTwo,
'targetOne': criticTargetNet,
'targetTwo': criticTargetNetTwo
}
optimizers = {
'actor': actorOptimizer,
'criticOne': criticOptimizer,
'criticTwo': criticOptimizerTwo
}
agent = TD3MultiStageUnit(config, actorNets, criticNets, env, optimizers,
torch.nn.MSELoss(reduction='mean'),
netParameter['n_output'])
agents.append(agent)
netParameter['n_output'] = N_A[1]
netParameter['n_hidden'] = [128]
policyNet = MultiLayerNetRegression(netParameter['n_feature'],
netParameter['n_hidden'],
netParameter['n_output'])
targetNet = MultiLayerNetRegression(netParameter['n_feature'],
netParameter['n_hidden'],
netParameter['n_output'])
optimizer = optim.Adam(policyNet.parameters(), lr=config['learningRate'])