-
Notifications
You must be signed in to change notification settings - Fork 0
/
experiment_config.yml
102 lines (93 loc) · 3.75 KB
/
experiment_config.yml
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
101
102
# If we should run on wandb
wandb:
use_wandb: True
entity: ''
project: ''
# path to save generated matrices to avoid needing to regen them
save_path_generated_matrices: './generated_matrices/'
# path to save csv file exported from wandb which provides templates by which to recover generated matrices if they are not present/available.
template_file: 'prior_lookup.csv'
datasets:
cifar10:
# Variants of CIFAR do not require you to download the zip yourself, just indicate a new root and they will automatically download
root_path: '~/cifar10/'
classes: 10
cifar3:
# Variants of CIFAR do not require you to download the zip yourself, just indicate a new root and they will automatically download
root_path: '~/cifar10/'
classes: 3
cifar20:
root_path: '~/cifar100/'
classes: 20
imagenet:
# ImageNet50 must be downloaded on your own. Please indicate the root directory of the dataset.
root_path: '~/imagenet/'
classes: 50
fg2:
# FieldGuide2 must be installed on your own. Please indicate the root directory of the dataset.
root_path: '~/fieldguide2/'
classes: 2
fg28:
# FieldGuide28 must be installed on your own. Please indicate the root directory of the dataset.
root_path: '~/fieldguide28/'
classes: 28
# use flag --replace_experiments_with_paper_main <dataset_name> to replace other experiments with these specially-defined experiments
paper_replication_experiments:
alphas: [0.5, 3, 10]
random_seed_waves:
0:
data_generation_seed: [236648222, 479046732, 960276699, 750507101, 568466179]
model_stochasticity_seed: [664632, 798678, 234261, 7845, 12361]
1:
data_generation_seed: [4, 23, 623, 23423, 66]
model_stochasticity_seed: [9802394, 20342394, 1329582, 575, 2352]
2:
data_generation_seed: [268773, 9947296, 76383, 2234, 12839]
model_stochasticity_seed: [263417, 568546, 88798, 3467834, 875]
3:
data_generation_seed: [78496, 692, 7766739, 1112, 951]
model_stochasticity_seed: [252680230, 667896448, 428142901, 118636565, 253807063]
4:
data_generation_seed: [89234, 8847675, 123456786, 93939, 659256]
model_stochasticity_seed: [434007571, 580817424, 3717700, 957555157, 250683833]
datasets:
cifar10:
domains: [10,15,20,25]
max_condition_numbers: [4,4,8]
cifar20:
domains: [20,25,30]
max_condition_numbers: [8,12,20]
imagenet:
domains: [50, 60]
max_condition_numbers: [200,205,210]
fg2:
domains: [10,7,5,3,2]
max_condition_numbers: [3,5,7]
fg28:
domains: [47, 42, 37, 28]
max_condition_numbers: [12, 20, 28]
# List of all experiments to conduct. If an experiment is commented out, it will not be conducted.
# Note: this is not an exhaustive list of the experiments conducted in our paper.
# Instead, this is designed to be a template for conducting a user's own experiments by modifying the following list.
experiments:
- dataset_settings:
dataset: 'cifar10'
#7636
class_prior_generation:
# domains must be at least the number of classes
domains: 10
# alpha must be positive
alpha: 0.5
# max cond must be positive. If it is too small, it is unlikely that a valid matrix could be generated.
max_condition_number: 4
class_prior_estimator: 'cluster_nmf'
data_generation_seed: 4
model_stochasticity_seed: 219
estimate_prior_valid_train: True
# if true, we'll use both valid and train data for clustering. If false, we'll use valid only.
retrain: False
# If true, we'll always retrain models. If false, we'll load models if they exist or retrain otherwise.
approaches:
# options: ['ddfa', 'ddfa_scan']. If DDFA_SCAN is selected, the SCAN baseline will also be computed.
# - 'ddfa'
- 'ddfa_scan'