def __init__(self, alpha, epsilon, gamma, output_dim, hidden_units,
input_dim, num_actions):
self.alpha = alpha
self.epsilon = epsilon
self.gamma = gamma
self.curr_state = np.array([None, None])
self.curr_action = 0
self.q = nn.SingleLayerNN(output_dim, hidden_units, input_dim,
nn.LinearActivation(),
nn.SigmoidActivation())
self.num_actions = num_actions
def __init__(self,
policy_units,
value_units,
state_vector_len,
num_actions,
actor_stepsize,
critic_stepsize,
lambda_pi,
lambda_v,
gamma,
cov=1,
clip_grads=False,
normalize=False):
NeuralACAgent.__init__(self, policy_units, value_units,
nn.LinearActivation(), nn.SigmoidActivation(),
nn.LinearActivation(), nn.SigmoidActivation(),
state_vector_len, 1, num_actions,
actor_stepsize, critic_stepsize, lambda_pi,
lambda_v, gamma, cov, cov, clip_grads,
normalize)
def __init__(self,
num_target_policy_hidden_units,
state_vector_len,
num_actions,
actor_stepsize,
critic_stepsize,
lambda_pi,
lambda_v,
gamma,
cov=1.0,
clip_grads=False,
clip_rho=True):
action_vector_len = 1
NNPolicyLinearOPACAgent.__init__(self, num_target_policy_hidden_units,
nn.LinearActivation(),
nn.SigmoidActivation(),
state_vector_len, action_vector_len,
num_actions, actor_stepsize,
critic_stepsize, lambda_pi, lambda_v,
gamma, cov, clip_grads, clip_rho)