def predict_std(self, state, deterministic_=True):
# states = np.zeros((self._batch_size, self._state_length), dtype=self._settings['float_type'])
# states[0, ...] = state
"""
if ( ('disable_parameter_scaling' in self._settings) and (self._settings['disable_parameter_scaling'])):
pass
else:
"""
state = norm_state(state, self._state_bounds)
state = np.array(state, dtype=self._settings['float_type'])
self._model.setStates(state)
# action_ = lasagne.layers.get_output(self._model.getActorNetwork(), state, deterministic=deterministic_).mean()
# action_ = scale_action(self._q_action()[0], self._action_bounds)
# if deterministic_:
# action_std = scale_action(self._q_action_std()[0], self._action_bounds)
if (('disable_parameter_scaling' in self._settings)
and (self._settings['disable_parameter_scaling'])):
action_std = self._q_action_std()
# action_std = self._q_action_std()[0] * (action_bound_std(self._action_bounds))
else:
action_std = self._q_action_std() * (action_bound_std(
self._action_bounds))
# else:
# action_ = scale_action(self._q_action()[0], self._action_bounds)
# action_ = q_valsActA[0]
return action_std
def predict_std(self, state, deterministic_=True):
state = norm_state(state, self._state_bounds)
state = np.array(state, dtype=self._settings['float_type'])
# action_std = self._model.getActorNetwork().predict(state, batch_size=1)[:,self._action_length:] * (action_bound_std(self._action_bounds))
action_std = self._q_action_std([state])[0] * action_bound_std(
self._action_bounds)
# print ("Policy std: ", repr(action_std))
return action_std
def predict_std(self, state, deterministic_=True):
state = norm_state(state, self._state_bounds)
state = np.array(state, dtype=self._settings['float_type'])
self._model.setStates(state)
if (('disable_parameter_scaling' in self._settings)
and (self._settings['disable_parameter_scaling'])):
action_std = self._q_action_std()[0]
else:
action_std = self._q_action_std()[0] * (action_bound_std(
self._action_bounds))
return action_std
def predict_std(self, state, deterministic_=True):
state = norm_state(state, self._state_bounds)
state = np.array(state, dtype=self._settings['float_type'])
self._model.setStates(state)
if (('disable_parameter_scaling' in self._settings)
and (self._settings['disable_parameter_scaling'])):
action_std = self._model.getActorNetwork().predict(
state, batch_size=1)[:, self._action_length:]
# action_std = self._q_action_std()[0] * (action_bound_std(self._action_bounds))
else:
action_std = self._model.getActorNetwork().predict(
state, batch_size=1)[:, self._action_length:] * (
action_bound_std(self._action_bounds))
return action_std