-
Notifications
You must be signed in to change notification settings - Fork 6
/
config.yaml
63 lines (57 loc) · 1.4 KB
/
config.yaml
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
device: 'cpu' # make sure to use `cuda` on GPU servers
is_learning: True
is_testing: True
test: False
folder_location: './baseline/'
folder_name: 'helicopter_env_test'
network_path: 'weights.pt'
batch: False
seed: 123
dataset_size: 1000
# learning_type should be regular to train a baseline
learning_type: 'regular'
# minimum_count corresponds to n_wedge in the paper. Set to 0 for vanilla DQN.
minimum_count: 0.0
# epsilon_count is the epsilon value used for soft spibb
epsilon_soft: 1.0
# kappa corresponds to the kappa parameter in ramdp. Set to 0 for vanilla DQN.
kappa: 0.003
# Parameters of the environment set to the ones used in the experiment
domain: 'helicopter'
helicopter_time_step: 0.1
helicopter_size: 3
helicopter_a_max: 1
helicopter_v_max: 1
helicopter_noise: 0.025
helicopter_noise_v: 0.05
noise_factor: 1
helicopter_log: False
num_experiments: 1
num_epochs: 20
passes_on_dataset: 2000
steps_per_test: 10000
episode_max_len: 100
max_start_nullops: 0
steps_per_epoch: 10000
extra_stochasticity: 0.
epsilon: 1
annealing: True
final_epsilon: 0
test_epsilon: 0
annealing_start: 0
annealing_steps: 2000000
ddqn: True
network_size: 'dense' # `large`=nips paper model, `nature`=nature paper model
gamma: .9
learning_rate: 0.01
minibatch_size: 32
update_freq: 10
state_shape: [4]
nb_actions: 9
history_len: 1
replay_max_size: 1000000
replay_min_size: 1
learning_frequency: 10
action_dim: 1
reward_dim: 1
normalize: 1.