-
Notifications
You must be signed in to change notification settings - Fork 16
/
run_bbb_exp.py
100 lines (87 loc) · 3.91 KB
/
run_bbb_exp.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
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
import experiments
import networks
if __name__ == '__main__':
import argparse
parser = argparse.ArgumentParser(
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument('--logdir', '-p', help='tb directory',
default='/vol/biomedic2/np716/bbh_nips/mnist/bbb/')
parser.add_argument('--experiment', '-x', help='tb directory',
default='test')
parser.add_argument('--seed', '-s', help='seed',
default=42, type=int)
parser.add_argument('--epochs', '-e', help='tb directory',
default=5, type=int)
parser.add_argument('--output_mc', '-m', help='', default=False,
action='store_true')
parser.add_argument('--annealing', '-a', help='', default=False,
action='store_true')
parser.add_argument('--random_weights', '-r', help='', default=0, type=int)
parser.add_argument('--lr', '-d', help='', default=0.001, type=float)
parser.add_argument('--prior_scale', help='', default=1.,
type=float)
parser.add_argument('--cuda', '-c', default='0')
parser.add_argument('--opt', '-o', help='', default='adam',
choices=['adam', 'rms'])
parser.add_argument('--align_noise', help='', default=False,
action='store_true')
args = parser.parse_args()
import os
import tensorflow as tf
os.environ['CUDA_VISIBLE_DEVICES'] = args.cuda
config = {}
config['logdir'] = os.path.join(args.logdir, args.experiment)
config['seed'] = args.seed
config['random_weights'] = args.random_weights
config['num_samples'] = 5
config['annealing'] = args.annealing
config['learning_rate'] = args.lr
config['annealing_epoch_start'] = 5
config['annealing_epoch_length'] = 15
config['disc_units'] = [64, 64]
config['disc_pretrain'] = 100
config['disc_train'] = 1
config['epochs'] = args.epochs
config['prior_scale'] = args.prior_scale
config['optimiser'] = 'adam'
config['mod'] = 'bbb'
config['args'] = str(args)
tf.reset_default_graph()
config['experiment'] = 'bbb_analytical'
config['args'] = str(args)
ops = networks.get_bbb_mnist({}, init_var=-9.,
prior_scale=args.prior_scale,
aligned_noise=args.align_noise)
experiments.run_analytical_experiment(ops, config)
tf.reset_default_graph()
config['experiment'] = 'bbb_klapprox'
config['full_kernel'] = False
config['args'] = str(args)
ops = networks.get_bbb_mnist({}, init_var=-9.,
prior_scale=args.prior_scale,
aligned_noise=args.align_noise)
experiments.run_klapprox_experiment(ops, config)
tf.reset_default_graph()
config['experiment'] = 'bbb_klapprox_fullkernel'
config['full_kernel'] = True
config['args'] = str(args)
ops = networks.get_bbb_mnist({}, init_var=-9.,
prior_scale=args.prior_scale,
aligned_noise=args.align_noise)
experiments.run_klapprox_experiment(ops, config)
tf.reset_default_graph()
config['experiment'] = 'bbb_disc'
config['args'] = str(args)
ops = networks.get_bbb_mnist({}, init_var=-9.,
prior_scale=args.prior_scale,
aligned_noise=args.align_noise)
experiments.run_disc_experiment(ops, config)
if config['random_weights'] > 0:
config['random_weights'] = 0
tf.reset_default_graph()
config['experiment'] = 'bbb_klapprox_r0'
config['args'] = str(args)
ops = networks.get_bbb_mnist({}, init_var=-9.,
prior_scale=args.prior_scale,
aligned_noise=args.align_noise)
experiments.run_klapprox_experiment(ops, config)