#max_step - number of steps within an episode
Experiments = 10
Experiments_All_Rewards = np.zeros(shape=(max_n_ep, Experiments))
lbda = 1
perdatapoint = 0
prior = log_normal
coupling = 4
for e in range(Experiments):
print("Experiment", e)
value_function = BHN_Q_Network(lbda=lbda,
perdatapoint=perdatapoint,
prior=prior,
coupling=coupling)
epsilon = 0.1
#decay rate for the temperature parameter
discount = 0.9
agent = AgentEpsGreedy(n_actions=n_actions,
value_function_model=value_function,
state_dim=state_dim,
batch_size=batch_size,
eps=epsilon)
memory = ReplayMemory(max_size=100000)
loss_per_ep = []
w1_m_per_ep = []
for e in range(Experiments):
print ("Experiment", e)
weight_shapes = [(512, 256), (256, 2)]
coupling_dim = 256
value_function = BHN_Q_Network(lbda=lbda, perdatapoint=perdatapoint, prior=prior, coupling=coupling, wn=WN,
weight_shapes=weight_shapes,
coupling_dim=coupling_dim)
epsilon = 0.1
#decay rate for the temperature parameter
discount = 0.9
agent = AgentEpsGreedy(n_actions=n_actions, value_function_model=value_function, state_dim=state_dim, batch_size=batch_size, eps=epsilon)
memory = ReplayMemory(max_size=100000)
loss_per_ep = []
w1_m_per_ep = []
w2_m_per_ep = []
w3_m_per_ep = []