OSHeDA/utils.py at main · AIMedLab/OSHeDA · GitHub

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
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
import torch
import numpy as np
import os
import pickle
import json
import scipy.io as sio


def save_model(model, model_dir, mode, setting, seed, svc=False):
    if svc:
        outfile = '%s_%d_%s.pkl' % (setting, seed, mode)
        with open(os.path.join(model_dir, outfile), 'wb') as f:
            pickle.dump({'model': model}, f)
        return
    outfile = '%s_%d_%s.ckpt' % (setting, seed, mode)
    torch.save({'model_state_dict': model.state_dict()},
               os.path.join(model_dir, outfile))


def load_model(model, hparams, hparams_seed, seed):
    if 'mode' in hparams.keys() and hparams['mode'] == 'dist_shift':
        ckpt_name = '%s_%s_%s_%s_%d_%d.ckpt' % (hparams['model'], hparams['src_dataset'], hparams['dst_dataset'], hparams['sensitive_attribute'], hparams_seed, seed)
    else:
        ckpt_name = '%s_%s_%s_%d_%d.ckpt' % (hparams['model'], hparams['dataset'], hparams['sensitive_attribute'], hparams_seed, seed)

    checkpoint = torch.load(os.path.join(hparams['model_dir'], ckpt_name), map_location=hparams['device'])
    if hparams['model'] == 'GroupDRO':
        model.q = (checkpoint['model_state_dict']['q'])
    model.load_state_dict(checkpoint['model_state_dict'], strict=True)
    model.to(hparams['device'])
    return model


def save_result(score_val, score_test, score_dir, setting, seed):
    outfile = '%s_%d.pkl' % (setting, seed)
    with open(os.path.join(score_dir, outfile), 'wb') as f:
        pickle.dump({'val': score_val, 'test': score_test}, f)


def generate_config():
    configuration = {}
    for dataset in ['Wikipedia', 'OfficeCaltech', 'NUSIMAGE', 'MRT', 'ImageCLEF', 'BT', 'PTBXL']:
        if dataset == 'Wikipedia': # 2 settings
            data_path = os.path.join('datasets', dataset)
            labeled_size = 5
            domains = ['Image_BT', 'TEXT_BB']
            for source in domains:
                for target in domains:
                    if source != target:
                        config = {'src_dir': os.path.join(data_path, 'L5', 'T_%s_L5.mat' % source),
                                'tgt_dir': os.path.join(data_path, 'L5', 'S_%s_L5.mat' % target),
                                'unk_idx': 5, 'n_labels': 6, 'lamda': 0.5}
                        tgt_data = sio.loadmat(config['tgt_dir'])
                        tgt_u_data = tgt_data['testing_labels'][0][0].reshape(-1) - 1
                        config['lamda'] = (tgt_u_data >= config['unk_idx']).sum() / len(tgt_u_data)
                        configuration['%s_%s_%s_%d' % (dataset, source, target, labeled_size)] = config
        elif dataset == 'OfficeCaltech': # 18 settings
            data_path = os.path.join('datasets', dataset)
            labeled_size = 3
            domains = ['amazon', 'Caltech10', 'webcam', 'dslr']
            features = ['DeCAF6', 'SURF_L10']
            for source in domains[:-1]:
                for target in domains:
                    for source_feature in features:
                        for target_feature in features:
                            if source != target and source_feature != target_feature:
                                config = {'src_dir': os.path.join(data_path, 'randsource_%s_%s.mat' % (source, source_feature)),
                                        'tgt_dir': os.path.join(data_path, 'randtarget_%s_%s.mat' % (target, target_feature)),
                                        'unk_idx': 5, 'n_labels': 6, 'lamda': 0.5}
                                tgt_data = sio.loadmat(config['tgt_dir'])
                                tgt_u_data = tgt_data['testing_labels'][0][0].reshape(-1) - 1
                                config['lamda'] = (tgt_u_data >= config['unk_idx']).sum() / len(tgt_u_data)
                                configuration['%s_%s_%s_%s_%s_%d' % (dataset, source, source_feature, target, target_feature, labeled_size)] = config
        elif dataset == 'NUSIMAGE': # 1 setting
            data_path = os.path.join('datasets', dataset)
            labeled_size = 3
            source = 'nustag_hist'
            target = 'imagenet_decaf'
            config = {'src_dir': os.path.join(data_path, '%s.mat' % source),
                    'tgt_dir': os.path.join(data_path, '%s.mat' % target),
                    'unk_idx': 4, 'n_labels': 5, 'lamda': 0.5}
            configuration['%s_%s_%s_%d' % (dataset, source, target, labeled_size)] = config
        elif dataset == 'MRT': # 4 settings
            data_path = os.path.join('datasets', dataset)
            labeled_size = 20
            sources = ['english', 'french', 'german', 'italian']
            target = 'spanish'
            for source in sources:
                config = {'src_dir': os.path.join(data_path, '%s_tfidf.mat' % source),
                        'tgt_dir': os.path.join(data_path, '%s_tfidf.mat' % target),
                        'unk_idx': 3, 'n_labels': 4, 'lamda': 0.5}
                configuration['%s_%s_tfidf_%s_tfidf_%d' % (dataset, source, target, labeled_size)] = config
        elif dataset == 'ImageCLEF': # 24 settings
            data_path = os.path.join('datasets', dataset)
            labeled_size = 3
            domains = ['b', 'c', 'i', 'p']
            features = ['r', 'v']
            for source in domains:
                for target in domains:
                    for source_feature in features:
                        for target_feature in features:
                            if source != target and source_feature != target_feature:
                                config = {'src_dir': os.path.join(data_path, '%s%s%s%ss.mat' % (source, source_feature, target, target_feature)),
                                        'tgt_dir': os.path.join(data_path, '%s%s%s%st.mat' % (source, source_feature, target, target_feature)),
                                        'unk_idx': 6, 'n_labels': 7, 'lamda': 0.5}
                                tgt_data = sio.loadmat(config['tgt_dir'])
                                tgt_u_data = tgt_data['testing_labels'][0][0].reshape(-1) - 1
                                config['lamda'] = (tgt_u_data >= config['unk_idx']).sum() / len(tgt_u_data)
                                configuration['%s_%s_%s_%s_%s_%d' % (dataset, source, source_feature, target, target_feature, labeled_size)] = config
        elif dataset == 'BT':
            data_path = os.path.join('datasets', dataset)
            labeled_size_list = [1, 3, 5]
            source = 'Source8'
            target = 'CIFAR8'
            features = [50, 101]
            for labeled_size in labeled_size_list:
                for source_feature in features:
                    for target_feature in features:
                        if source_feature != target_feature:
                            config = {'src_dir': os.path.join(data_path, 'Data_random_L%d' % labeled_size, '%s_%d_L%d.mat' % (source, source_feature, labeled_size)),
                                    'tgt_dir': os.path.join(data_path, 'Data_random_L%d' % labeled_size, '%s_%d_L%d.mat' % (target, target_feature, labeled_size)),
                                    'unk_idx': 4, 'n_labels': 5, 'lamda': 0.5}
                            tgt_data = sio.loadmat(config['tgt_dir'])
                            tgt_u_data = tgt_data['testing_labels'][0][0].reshape(-1) - 1
                            config['lamda'] = (tgt_u_data >= config['unk_idx']).sum() / len(tgt_u_data)
                            configuration['%s_%s_%d_%s_%d_%d' % (dataset, source, source_feature, target, target_feature, labeled_size)] = config
        elif dataset == 'PTBXL':
            # data_path = os.path.join('datasets', dataset)
            data_path = os.path.join('ptb-xl')
            labeled_size = 20
            source, target = 'ECG', 'ECG'
            source_feature, target_feature = 'signal', 'image'
            config = {'data_dir': data_path, 'unk_idx': 4, 'n_labels': 5, 'lamda': 0.13081269013472405}
            configuration['%s_%s_%s_%s_%s_%d' % (dataset, source, source_feature, target, target_feature, labeled_size)] = config
    with open("configuration.json", "w") as outfile:
        json.dump(configuration, outfile, indent=4)
    return configuration


def get_config():
    with open("configuration.json", "r") as infile:
        configuration = json.load(infile)
    return configuration


# generate_config()