def __init__(self, gamma=0.95, threshold=0):
ValueBasedLearner.__init__(self)
self.num_features = 5
self.num_actions = 1
self.sigma = 1.0
self.kern_c = 10
self.kern_sigma = 0.5
self.thresh = threshold
self.gamma = gamma
self.laststate = None
self.lastaction = None
self.lastreward = None
self.state_dict = None
self.cum_reward = np.array([])
self.u_tilde = np.array([])
self.C_tilde = np.array([[]])
self.d = 0.0
self.v_inv = 0.0
self.c_tild = np.array([])
self.dataset = None
#self.g=np.array([])
self.K_inv = np.array([[]])
def __init__(self, alpha=0.5, gamma=0.99):
ValueBasedLearner.__init__(self)
self.alpha = alpha
self.gamma = gamma
self.laststate = None
self.lastaction = None
def __init__(self, alpha=0.5, gamma=0.99):
ValueBasedLearner.__init__(self)
self.alpha = alpha
self.gamma = gamma
self.laststate = None
self.lastaction = None
def __init__(self, alpha=0.5, gamma=0.99, neg_reward=False):
ValueBasedLearner.__init__(self)
self.alpha = alpha
self.gamma = gamma
self.neg_reward = neg_reward
self.laststate = None
self.lastactions = None
def __init__(self, maxEpochs=20, indexOfAgent=None,):
ValueBasedLearner.__init__(self)
self.gamma = 0.9
self.maxEpochs = maxEpochs
#
self.ownerAgentProperties["requireOtherAgentsState"]=False
self.ownerAgentProperties["requireJointAction"]=False
self.ownerAgentProperties["requireJointReward"]=False
self.isFirstLerning=True
def __init__(self, alpha=1.0, w=1.0, gamma=0.99, iters=10):
ValueBasedLearner.__init__(self)
self.alpha = alpha # step scale
self.w = w # learning rate
self.gamma = gamma # temporal discount
self.iters = 10 # number of times to propagate value changes
self.step = 0
def __init__(self, num_actions, num_features, **kwargs):
ValueBasedLearner.__init__(self)
setAllArgs(self, kwargs)
self.explorer = None
self.num_actions = num_actions
self.num_features = num_features
if self.randomInit:
self._theta = randn(self.num_actions, self.num_features) / 10.
else:
self._theta = zeros((self.num_actions, self.num_features))
self._additionalInit()
self._behaviorPolicy = self._boltzmannPolicy
self.reset()
def __init__(self, num_actions, num_features, **kwargs):
ValueBasedLearner.__init__(self)
setAllArgs(self, kwargs)
self.explorer = None
self.num_actions = num_actions
self.num_features = num_features
if self.randomInit:
self._theta = randn(self.num_actions, self.num_features) / 10.
else:
self._theta = zeros((self.num_actions, self.num_features))
self._additionalInit()
self._behaviorPolicy = self._boltzmannPolicy
self.reset()
def __init__(
self,
maxEpochs=20,
indexOfAgent=None,
):
ValueBasedLearner.__init__(self)
self.gamma = 0.9
self.maxEpochs = maxEpochs
#
self.ownerAgentProperties["requireOtherAgentsState"] = False
self.ownerAgentProperties["requireJointAction"] = False
self.ownerAgentProperties["requireJointReward"] = False
self.isFirstLerning = True
def __init__(self, gamma=0.99):
ValueBasedLearner.__init__(self)
self.gamma = gamma
self.laststate = None
self.lastaction = None
self.num_features = 2
self.num_actions = 1
self.kern_c = 10
self.covariance_mat = np.array([[]])
self.inv = np.array([])
self.state_dict = None
self.cum_reward = np.array([])
self.H = []
self.kern_sigma = 0.2
self.dataset = None
self.sigma = 1
def reset(self):
ValueBasedLearner.reset(self)
self._callcount = 0
self.newEpisode()
def __init__(self, indexOfAgent=None, **kwargs):
ValueBasedLearner.__init__(self)
self.indexOfAgent = indexOfAgent
def reset(self):
ValueBasedLearner.reset(self)
self.laststate = None
self.lastaction = None
def __init__(self, maxEpochs=20):
ValueBasedLearner.__init__(self)
self.gamma = 0.9
self.maxEpochs = maxEpochs
def __init__(self, alpha=0.5):
ValueBasedLearner.__init__(self)
self.alpha = alpha
self.gamma = config.getf('gammaDiscountReward')
self.netManager = CaffeMultiLayerPerceptronManagement(config.get('networkDir'))
def reset(self):
ValueBasedLearner.reset(self)
self._callcount = 0
self.newEpisode()
def newEpisode(self):
ValueBasedLearner.newEpisode(self)
self._callcount += 1
self.learningRate *= ((self.learningRateDecay + self._callcount) /
(self.learningRateDecay + self._callcount + 1.))
def __init__(self):
ValueBasedLearner.__init__(self)
self.gamma = 0.9
def __init__(self, alpha=0.5):
ValueBasedLearner.__init__(self)
self.alpha = alpha
self.gamma = config.getf('gammaDiscountReward')
self.netManager = CaffeMultiLayerPerceptronManagement(
config.get('networkDir'))
def __init__(self, indexOfAgent=None, **kwargs):
ValueBasedLearner.__init__(self)
self.indexOfAgent=indexOfAgent
def __init__(self, alpha=0.5, gamma=0.99):
ValueBasedLearner.__init__(self)
self.alpha = alpha
self.gamma = gamma
self.netManager = cnm.CaffeConvNetManagement(config.get('networkDir'))
def __init__(self, maxEpochs=20):
ValueBasedLearner.__init__(self)
self.gamma = 0.9
self.maxEpochs = maxEpochs
def newEpisode(self):
ValueBasedLearner.newEpisode(self)
self._callcount += 1
self.learningRate *= ((self.learningRateDecay + self._callcount)
/ (self.learningRateDecay + self._callcount + 1.))
epis=10000/batch #number of roleouts
numbExp=1 #number of experiments
env = None
for runs in range(numbExp):
# create environment
#Options: Bool(OpenGL), Bool(Realtime simu. while client is connected), ServerIP(default:localhost), Port(default:21560)
if env != None: env.closeSocket()
env = Tetra2Environment()
# create task
task = WalkForwardTask(env)
# create automatic programmer module for the robot (it will be received already created from the robot)
channels_setup = [(1 if i!=2 else 0,1, (0,255), 0) for i in range(5)]
programmer = Programmer(steps = 1000, channels_setup = channels_setup, types_subset = [1]) #, 2, 3])
learner = ValueBasedLearner()
agent = TetrapodAgent(programmer, learner)
# create the experiment
experiment = EpisodicExperiment(task, agent)
for _ in range(epis):
experiment.doEpisodes(batch)
agent.learn(total_reward = task.getTotalReward())
