def __init__(self,
dU,
obs_tensor,
act_op,
feat_op,
var,
sess,
device_string,
copy_param_scope=None):
Policy.__init__(self)
self.dU = dU
self.obs_tensor = obs_tensor
self.act_op = act_op
self.feat_op = feat_op
self.sess = sess
self.device_string = device_string
self.chol_pol_covar = np.diag(np.sqrt(var))
self.scale = None # must be set from elsewhere based on observations
self.bias = None
self.x_idx = None
if copy_param_scope:
self.copy_params = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES,
scope=copy_param_scope)
self.copy_params_assign_placeholders = [
tf.placeholder(tf.float32, shape=param.get_shape())
for param in self.copy_params
]
self.copy_params_assign_ops = [
tf.assign(self.copy_params[i],
self.copy_params_assign_placeholders[i])
for i in range(len(self.copy_params))
]
def __init__(self, test_net, deploy_net, var):
Policy.__init__(self)
self.net = test_net
self.deploy_net = deploy_net
self.chol_pol_covar = np.diag(np.sqrt(var))
self.scale = None # must be set from elsewhere based on observations
self.bias = None
def __init__(self, agent, learning_rate, cond, noise_var = None):
Policy.__init__(self)
self.agent = agent
self.learning_rate = learning_rate
if noise_var is not None:
self.sqrt_noise_var = np.sqrt(noise_var)
self.cond = cond # cond, not m
def __init__(self, algorithm, history_length):
Policy.__init__(self)
self.dU = algorithm.dU # This should be in super class
self.algorithm = algorithm
self.X = np.empty([algorithm.T, algorithm.dX])
self.U = np.empty([algorithm.T, algorithm.dU])
self.history_length = history_length
def __init__(self, dU, dX, var, gcm):
Policy.__init__(self)
self.gcm = gcm
self.dU = dU
self.dX = dX
self.chol_pol_covar = np.diag(np.sqrt(var))
self.scale = None # must be set from elsewhere based on observations
self.bias = None
self.x_idx = None
def __init__(self, dU, obs_tensor, act_op, var, sess, device_string):
Policy.__init__(self)
self.dU = dU
self.obs_tensor = obs_tensor
self.act_op = act_op
self.sess = sess
self.device_string = device_string
self.chol_pol_covar = np.diag(np.sqrt(var))
self.scale = None # must be set from elsewhere based on observations
self.bias = None
def __init__(self, dU, obs_tensor, act_op, var, sess, device_string):
Policy.__init__(self)
self.dU = dU
self.obs_tensor = obs_tensor
self.act_op = act_op
self.sess = sess
self.device_string = device_string
self.chol_pol_covar = np.diag(np.sqrt(var))
self.scale = None # must be set from elsewhere based on observations
self.bias = None
def __init__(self, K, k, pol_covar, chol_pol_covar, inv_pol_covar):
Policy.__init__(self)
# Assume K has the correct shape, and make sure others match.
self.T = K.shape[0]
self.dU = K.shape[1]
self.dX = K.shape[2]
check_shape(k, (self.T, self.dU))
check_shape(pol_covar, (self.T, self.dU, self.dU))
check_shape(chol_pol_covar, (self.T, self.dU, self.dU))
check_shape(inv_pol_covar, (self.T, self.dU, self.dU))
self.K = K
self.k = k
self.pol_covar = pol_covar
self.chol_pol_covar = chol_pol_covar
self.inv_pol_covar = inv_pol_covar
def __init__(self, dU, obs_tensor, act_op, feat_op, var, sess, device_string, copy_param_scope=None):
Policy.__init__(self)
self.dU = dU
self.obs_tensor = obs_tensor
self.act_op = act_op
self.feat_op = feat_op
self.sess = sess
self.device_string = device_string
self.chol_pol_covar = np.diag(np.sqrt(var))
self.scale = None # must be set from elsewhere based on observations
self.bias = None
self.x_idx = None
if copy_param_scope:
self.copy_params = tf.get_collection(tf.GraphKeys.VARIABLES, scope=copy_param_scope)
self.copy_params_assign_placeholders = [tf.placeholder(tf.float32, shape=param.get_shape()) for
param in self.copy_params]
self.copy_params_assign_ops = [tf.assign(self.copy_params[i],
self.copy_params_assign_placeholders[i])
for i in range(len(self.copy_params))]
def __init__(self, dU):
Policy.__init__(self)
self.dU = dU
def __init__(self, const=0.5):
Policy.__init__(self)
self.const = const
self.adapt_sigma = CMAAdaptSigmaCSA()
def __init__(self, T=50):
Policy.__init__(self)
self.teacher = 0 #np.random.choice([0,1])
self.T = T
self.adapt_sigma = CMAAdaptSigmaCSA()
def __init__(self, test_net, deploy_net, var):
Policy.__init__(self)
self.net = test_net
self.deploy_net = deploy_net
self.chol_pol_covar = np.diag(np.sqrt(var))
