forked from nicklashansen/adaptive-learning-rate-schedule
-
Notifications
You must be signed in to change notification settings - Fork 0
/
baselines.py
70 lines (54 loc) · 2.5 KB
/
baselines.py
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
64
65
66
67
68
69
70
import numpy as np
import torch
import utils
if __name__ == '__main__':
args = utils.parse_args()
print(f'Running baseline for ALRS testing...\nArgs:\n{utils.args_to_str(args)}\n')
displayed_rendering_error = False
best_config = None
best_info_list = None
best_val_loss = np.inf
initial_lrs = [1e-1, 1e-2, 1e-3, 1e-4]
discount_steps = [10, 20, 50, 100]
discount_factors = [.99, .9, .88]
for initial_lr in initial_lrs:
for discount_step in discount_steps:
for discount_factor in discount_factors:
print(f'Initial LR: {initial_lr}\nDiscount step: {discount_step}\nDiscount factor: {discount_factor}')
args.initial_lr = initial_lr
env = utils.make_alrs_env(args, test=True, baseline=True)
env.reset()
done = False
global_step = 0
current_lr = initial_lr
info_list = []
while not done:
action, new_lr = utils.step_decay_action(current_lr, global_step, discount_step, discount_factor)
_, _, done, info = env.step(action)
global_step += args.update_freq
current_lr = new_lr
info_list.append(info)
try:
env.render()
except:
if not displayed_rendering_error:
displayed_rendering_error = True
print('Warning: device does not support rendering.')
val_loss = env.venv.envs[0].env.latest_end_val
print('Final validation loss:', val_loss)
if val_loss < best_val_loss:
best_config = {
'dataset': args.dataset,
'architecture': args.architecture,
'initial_lr': initial_lr,
'discount_step': discount_step,
'discount_factor': discount_factor,
'val_loss': val_loss,
'log_val_loss': np.log(val_loss)
}
best_info_list = info_list
best_val_loss = val_loss
print(f'Found best configuration:\n{best_config}')
filename = args.dataset+'_'+args.architecture
utils.dict_to_file(best_config, filename, path='data/baselines/')
utils.save_baseline(best_info_list, filename)