max_n_ep = 2000 #number of episodes
#max_step - number of steps within an episode
Experiments = 1
Experiments_All_Rewards = np.zeros(shape=(max_n_ep))
for e in range(Experiments):
value_function = ValueFunctionDQN_TEST_TRAIN_DROPOUT(state_dim=state_dim, n_actions=n_actions, batch_size=batch_size)
epsilon = 0.1
#decay rate for the temperature parameter
# discount = 0.9
agent = AgentEpsGreedy(n_actions=n_actions, value_function_model=value_function, eps=epsilon)
memory = ReplayMemory(max_size=100000)
loss_per_ep = []
w1_m_per_ep = []
w2_m_per_ep = []
w3_m_per_ep = []
total_reward = []
#at the start of episodes
#initialise the Q network with dropout probability 0.9
#set current model index M with p=0.9
dropout_probability = 0.9
ep = 0
avg_Rwd = -np.inf
episode_end_msg = 'loss={:2.10f}, w1_m={:3.1f}, w2_m={:3.1f}, w3_m={:3.1f}, total reward={}'
#max_step - number of steps within an episode
Experiments = 10
Experiments_All_Rewards = np.zeros(shape=(max_n_ep, Experiments))
for e in range(Experiments):
#using a smaller network for the value function
value_function = ValueFunctionDQN_TEST_TRAIN_DROPOUT(state_dim=state_dim,
n_actions=n_actions,
batch_size=batch_size)
#this is the temperature parameter in Boltzmann Distribution
tau_parameter = 1
agent = AgentEpsGreedy(n_actions=n_actions,
value_function_model=value_function,
eps=tau_parameter)
memory = ReplayMemory(max_size=100000)
print('Experiment Number ', e)
loss_per_ep = []
w1_m_per_ep = []
w2_m_per_ep = []
w3_m_per_ep = []
total_reward = []
ep = 0
avg_Rwd = -np.inf
episode_end_msg = 'loss={:2.10f}, w1_m={:3.1f}, w2_m={:3.1f}, w3_m={:3.1f}, total reward={}'