MountaincarActor.cnnt 804 Bytes
Newer Older
Nicola Gatto's avatar
Nicola Gatto committed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
configuration MountaincarActor {
    context : cpu

    learning_method : reinforcement
    rl_algorithm : ddpg-algorithm
    critic : mountaincar.agent.mountaincarCritic

    environment : gym {name : "MountainCarContinuous-v0"}

    num_episodes : 200
    discount_factor : 0.999
    num_max_steps : 1000
    training_interval : 1

    snapshot_interval : 20

    replay_memory : buffer{
Nicola Gatto's avatar
Nicola Gatto committed
18
19
        memory_size : 1000000
        sample_size : 64
Nicola Gatto's avatar
Nicola Gatto committed
20
21
    }

Nicola Gatto's avatar
Nicola Gatto committed
22
    strategy : ornstein_uhlenbeck{
Nicola Gatto's avatar
Nicola Gatto committed
23
24
25
26
        epsilon : 1.0
        min_epsilon : 0.01
        epsilon_decay_method: linear
        epsilon_decay : 0.01
Nicola Gatto's avatar
Nicola Gatto committed
27
28
29
30
31
32
33
        mu: (0.0)
        theta: (0.15)
        sigma: (0.3)
    }

    actor_optimizer : adam {
        learning_rate : 0.0001
Nicola Gatto's avatar
Nicola Gatto committed
34
35
    }

Nicola Gatto's avatar
Nicola Gatto committed
36
    critic_optimizer : adam {
Nicola Gatto's avatar
Nicola Gatto committed
37
38
39
        learning_rate : 0.001
    }
}