def make_agent(agent_type, env, num_clusters, use_extra_travel_penalty,
use_extra_bit, use_controller_dqn, use_intrinsic_timeout,
use_memory, memory_size, pretrain_controller):
if agent_type == 'dqn':
return dqn.DqnAgent(state_dims=[2],
num_actions=2) # env.action_space.n
elif agent_type == 'h_dqn':
meta_controller_state_fn, check_subgoal_fn, num_subgoals, subgoals = clustering.get_cluster_fn(
n_clusters=num_clusters, extra_bit=use_extra_bit)
return hierarchical_dqn.HierarchicalDqnAgent(
state_sizes=[num_subgoals, [2]],
agent_types=['tabular', 'network'],
subgoals=subgoals,
num_subgoals=num_subgoals,
num_primitive_actions=2, # env.action_space.n
meta_controller_state_fn=meta_controller_state_fn,
check_subgoal_fn=check_subgoal_fn,
use_extra_travel_penalty=use_extra_travel_penalty,
use_extra_bit_for_subgoal_center=use_extra_bit,
use_controller_dqn=use_controller_dqn,
use_intrinsic_timeout=use_intrinsic_timeout,
use_memory=use_memory,
memory_size=memory_size,
pretrain_controller=pretrain_controller)
def make_agent(agent_type, env):
if agent_type == 'dqn':
return dqn.DqnAgent(state_dims=[2],
num_actions=2) # env.action_space.n
elif agent_type == 'h_dqn':
meta_controller_state_fn, check_subgoal_fn, num_subgoals, subgoals = clustering.get_cluster_fn(
n_clusters=4, extra_bit=False)
return hierarchical_dqn.HierarchicalDqnAgent(
state_sizes=[[num_subgoals], 2],
subgoals=subgoals,
num_subgoals=num_subgoals,
num_primitive_actions=2, # env.action_space.n
meta_controller_state_fn=meta_controller_state_fn,
check_subgoal_fn=check_subgoal_fn)
def make_agent(agent_type, env, load=True):
if agent_type == 'dqn':
return dqn.DqnAgent(state_dims=[2], num_actions=env.action_space.n)
elif agent_type == 'h_dqn':
meta_controller_state_fn, check_subgoal_fn, num_subgoals = None, check_subgoal, 2
# subgoals = [\
# [-.7,-.2],
# [-1,0],
# [.5,.2],
# [ 1,0]
# ]
#clustering.get_cluster_fn(n_clusters=num_clusters, extra_bit=use_extra_bit)
return hierarchical_dqn.HierarchicalDqnAgent(
state_sizes=env.observation_space.shape,
subgoals=subgoals,
num_subgoals=num_subgoals,
num_primitive_actions=env.action_space.n,
meta_controller_state_fn=meta_controller_state_fn,
check_subgoal_fn=check_subgoal_fn,
load=load)
logger = logging.getLogger()
logger.setLevel(logging.INFO)
env = gym.make(pms.gameName)
# You provide the directory to write to (can be an existing
# directory, including one with existing data -- all monitor files
# will be namespaced). You can also dump to a tempdir if you'd
# like: tempfile.mkdtemp().
outdir = '/tmp/DQN-' + pms.gameName
env.monitor.start(outdir, force=True, seed=0)
# This declaration must go *after* the monitor call, since the
# monitor's seeding creates a new action_space instance with the
# appropriate pseudorandom number generator.
agent = dqn.DqnAgent(env.action_space)
tran = Tran(max_size=pms.bufferSize)
caffe.set_mode_gpu()
imageDim = np.array((pms.frameHeight, pms.frameWidth))
curFrame = np.zeros((pms.frameChannel, pms.frameHeight, pms.frameWidth))
nextFrame = np.zeros((pms.frameChannel, pms.frameHeight, pms.frameWidth))
testStep = 0
update_step = 0
for i in range(pms.episodeCount):
rgbImage = env.reset()
# env.render()
done = False
for j in range(pms.frameChannel):
logger.setLevel(logging.INFO)
env = gym.make(pms.gameName)
# You provide the directory to write to (can be an existing
# directory, including one with existing data -- all monitor files
# will be namespaced). You can also dump to a tempdir if you'd
# like: tempfile.mkdtemp().
outdir = '/tmp/DQN-Test-' + pms.gameName
env.monitor.start(outdir, force=True, seed=0)
# This declaration must go *after* the monitor call, since the
# monitor's seeding creates a new action_space instance with the
# appropriate pseudorandom number generator.
model = './models/action_iter_2000000.caffemodel'
agent = dqn.DqnAgent(env.action_space, model=model)
caffe.set_mode_gpu()
imageDim = np.array((pms.frameHeight, pms.frameWidth))
curFrame = np.zeros((pms.frameChannel, pms.frameHeight, pms.frameWidth))
nextFrame = np.zeros((pms.frameChannel, pms.frameHeight, pms.frameWidth))
testStep = 0
update_step = 0
for i in range(pms.episodeTestCount):
rgbImage = env.reset()
env.render()
done = False
rewardSum = 0
while (done == False):