criticTargetNet = deepcopy(criticNet)
actorOptimizer = optim.Adam(actorNet.parameters(),
lr=config['actorLearningRate'])
criticOptimizer = optim.Adam(criticNet.parameters(),
lr=config['criticLearningRate'])
actorNets = {'actor': actorNet, 'target': actorTargetNet}
criticNets = {'critic': criticNet, 'target': criticTargetNet}
optimizers = {'actor': actorOptimizer, 'critic': criticOptimizer}
agent = DDPGAgent(config,
actorNets,
criticNets,
env,
optimizers,
torch.nn.MSELoss(reduction='mean'),
N_A,
stateProcessor=stateProcessor,
experienceProcessor=experienceProcessor)
checkpoint = torch.load('../../Log/Epoch15000_checkpoint.pt')
agent.actorNet.load_state_dict(checkpoint['actorNet_state_dict'])
config['randomMoveFlag'] = True
config['dynamicInitialStateFlag'] = False
config['dynamicTargetFlag'] = False
config['currentState'] = [-20, -20, 1, 1, 0, 0]
config['currentState'] = [0, 0, 1, 1, 0, 0]
config['targetState'] = [15, 15, 25]
actorTargetNet = deepcopy(actorNet)
criticNet = Critic(netParameter['n_feature'] + N_A, netParameter['n_hidden'])
criticTargetNet = deepcopy(criticNet)
actorOptimizer = optim.Adam(actorNet.parameters(),
lr=config['actorLearningRate'])
criticOptimizer = optim.Adam(criticNet.parameters(),
lr=config['criticLearningRate'])
actorNets = {'actor': actorNet, 'target': actorTargetNet}
criticNets = {'critic': criticNet, 'target': criticTargetNet}
optimizers = {'actor': actorOptimizer, 'critic': criticOptimizer}
agent = DDPGAgent(config, actorNets, criticNets, env, optimizers,
torch.nn.MSELoss(reduction='mean'), N_A)
plotPolicyFlag = False
N = 100
if plotPolicyFlag:
phi = 0.0
xSet = np.linspace(-10, 10, N)
ySet = np.linspace(-10, 10, N)
policy = np.zeros((N, N))
value = np.zeros((N, N))
for i, x in enumerate(xSet):
for j, y in enumerate(ySet):
# x, y is the target position, (0, 0, 0) is the particle configuration
distance = np.array([x, y])
