def __init__(self, policy_params):
Policy.__init__(self, policy_params)
self.numvars = policy_params['numvars']
hsize = policy_params['hsize']
numlayers = policy_params['numlayers']
rowembeddim = policy_params['rowembed'] # row embedding
embeddeddim = policy_params['embed'] # attention embedding
#self.make_mlp_weights(policy_params['numvars']+1, embeddeddim, policy_params['hsize'], policy_params['numlayers'])
self.make_mlp_weights(rowembeddim, embeddeddim, policy_params['hsize'],
policy_params['numlayers'])
# embed
self.embedlayer = lstmlayer(nin=1, nh=rowembeddim)
self.rowembeddim = policy_params['rowembed']
self.embedoffset = self.embedlayer.get_weights().size
# build up filter
self.observation_filter = get_filter(policy_params['ob_filter'],
shape=(self.numvars + 1, ))
# for visualization
"""
self.baseobsdict = []
self.normalized_attentionmap = []
self.cutsdict = []
"""
self.t = 0
def __init__(self, policy_params):
Policy.__init__(self, policy_params)
self.numvars = policy_params['numvars']
self.weight = np.random.randn(self.numvars + 1) * 0.001
# build up filter
self.observation_filter = get_filter(policy_params['ob_filter'],
shape=(self.numvars + 1, ))
def __init__(self, policy_params):
self.numvars = policy_params['numvars']
self.weights = np.empty(0)
# a filter for updating statistics of the observations and normalizing inputs to the policies
#if False:
if True:
self.observation_filter = get_filter(policy_params['ob_filter'],
shape=(2, ))
self.update_filter = True
def __init__(self, policy_params):
Policy.__init__(self, policy_params)
self.numvars = policy_params['numvars']
hsize = policy_params['hsize']
numlayers = policy_params['numlayers']
self.make_mlp_weights(policy_params['numvars'] + 1, 1,
policy_params['hsize'],
policy_params['numlayers'])
# build up filter
self.observation_filter = get_filter(policy_params['ob_filter'],
shape=(self.numvars + 1, ))
def __init__(self, policy_params):
Policy.__init__(self, policy_params)
self.numvars = policy_params['numvars']
hsize = policy_params['hsize']
numlayers = policy_params['numlayers']
embeddeddim = policy_params['embed']
self.make_mlp_weights(policy_params['numvars'] + 1, embeddeddim,
policy_params['hsize'],
policy_params['numlayers'])
# build up filter
self.observation_filter = get_filter(policy_params['ob_filter'],
shape=(self.numvars + 1, ))
# for visualization
"""
self.baseobsdict = []
self.normalized_attentionmap = []
self.cutsdict = []
"""
self.t = 0