deep-CCA-tensorflow/utilities.py at main · alexanderdann/deep-CCA-tensorflow · 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
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
import tensorflow as tf
import numpy as np
import pandas as pd
import scipy.io as sio
from sklearn.utils import gen_batches
from sklearn.model_selection import StratifiedKFold
from sklearn.svm import LinearSVC as SVM
from sklearn.metrics import accuracy_score

"""
    Used to suppress the warning related to very big dataset in the SVM and track_test_progress
"""
from sklearn.utils._testing import ignore_warnings
from sklearn.exceptions import ConvergenceWarning


def batch_data(data, batch_size):
    channels, samples = data.shape
    num_batches = samples // batch_size

    tmp = np.zeros(shape=(num_batches, batch_size, channels), dtype=np.float32)
    for batch_idx, indeces in enumerate(gen_batches(samples, batch_size)):
        tmp[batch_idx] = data[:, indeces].T

    assert num_batches_1 == num_batches_2
    return tf.convert_to_tensor(tmp, dtype=tf.float32), num_batches


def prepare_data(data, batch_size):
    '''
        Used to make batches of data. Either batching data into multiple batches which makes
        use of 'batch_data()' (if-case) or just converts to tensor and adds one dimension (else-case)
    '''
    proc_data, batches = list(), dict()

    for view_idx, view_data in enumerate(data):
        tmp_data = dict()

        for key in ['train', 'validation', 'test']:
            if batch_size is not None:
                batch, num_batches = batch_data(view_data[key]['data'], batch_size)

            else:
                batch, num_batches = tf.convert_to_tensor(view_data[key]['data'].T, dtype=tf.float32)[None], 1

            tmp_data[key] = batch
            batches[key] = num_batches

        proc_data.append(tmp_data)

    return proc_data, batches


def load_mlsp_data():
    labels = pd.read_csv('mlsp-2014-mri/Train/train_labels.csv').to_numpy()[:, 1]
    labeled_FNC = pd.read_csv('mlsp-2014-mri/Train/train_FNC.csv').to_numpy()[:, 1:]
    labeled_SBM = pd.read_csv('mlsp-2014-mri/Train/train_SBM.csv').to_numpy()[:, 1:]

    unlabeled_FNC = pd.read_csv('mlsp-2014-mri/Test/test_FNC.csv').to_numpy()[:, 1:]
    unlabeled_SBM = pd.read_csv('mlsp-2014-mri/Test/test_SBM.csv').to_numpy()[:, 1:]


    data_FNC = list()
    data_SBM = list()

    for train_idx, tmp_idx in StratifiedKFold(n_splits=4).split(np.zeros(len(labels)), labels):
        for val_idx, test_idx in StratifiedKFold(n_splits=2).split(np.zeros(len(labels[tmp_idx])), labels[tmp_idx]):
            for modality, data in [('FNC', labeled_FNC), ('SBM', labeled_SBM)]:

                X_train, y_train = data[train_idx], labels[train_idx]
                X_test, y_test = data[tmp_idx][test_idx], labels[tmp_idx][test_idx]
                X_val, y_val = data[tmp_idx][val_idx], labels[tmp_idx][val_idx]

                if modality == 'FNC':
                    data_dict = {'unlabeled': {'data': unlabeled_FNC.T, 'labels': None},
                                 'train': {'data': X_train.T, 'labels': y_train},
                                 'validation': {'data': X_val.T, 'labels': y_val},
                                 'test': {'data': X_test.T, 'labels': y_test}}
                    data_FNC.append(data_dict)

                elif modality == 'SBM':
                    data_dict = {'unlabeled': {'data': unlabeled_SBM.T, 'labels': None},
                                 'train': {'data': X_train.T, 'labels': y_train},
                                 'validation': {'data': X_val.T, 'labels': y_val},
                                 'test': {'data': X_test.T, 'labels': y_test}}
                    data_SBM.append(data_dict)


    return data_FNC, data_SBM, labels


def compute_termination_score(train_data, train_labels, test_data, test_labels):
    assert train_data.shape[0] == train_labels.shape[0]
    assert test_data.shape[0] == test_labels.shape[0]
    svm_model = SVM(random_state=333)
    svm_model.fit(train_data.numpy(), train_labels)
    predictions = svm_model.predict(test_data.numpy())
    return accuracy_score(test_labels, predictions)


@ignore_warnings(category=ConvergenceWarning)
def track_test_progress_v2(model, data, test_labels, validation_labels):
    test_fy_1, test_fy_2 = model([data[0]['test'][0], data[1]['test'][0]])
    test_fy = tf.concat([test_fy_1, test_fy_2], axis=1)

    val_fy_1, val_fy_2 = model([data[0]['validation'][0], data[1]['validation'][0]])
    val_fy = tf.concat([val_fy_1, val_fy_2], axis=1)

    return compute_termination_score(val_fy, validation_labels, test_fy, test_labels)


@ignore_warnings(category=ConvergenceWarning)
def track_validation_progress(model, data, train_labels, validation_labels):
    train_fy_1, train_fy_2 = model([data[0]['train'][0], data[1]['train'][0]])
    train_fy = tf.concat([train_fy_1, train_fy_2], axis=1)

    val_fy_1, val_fy_2 = model([data[0]['validation'][0], data[1]['validation'][0]])
    val_fy = tf.concat([val_fy_1, val_fy_2], axis=1)

    return compute_termination_score(train_fy, train_labels, val_fy, validation_labels)


@ignore_warnings(category=ConvergenceWarning)
def track_test_progress(model, data, train_labels, validation_labels, test_labels):
    train_fy_1, train_fy_2 = model([data[0]['train'][0], data[1]['train'][0]])
    val_fy_1, val_fy_2 = model([data[0]['validation'][0], data[1]['validation'][0]])

    train_val_1 = tf.concat([train_fy_1, val_fy_1], axis=0)
    train_val_2 = tf.concat([train_fy_2, val_fy_2], axis=0)
    train_val_concat = tf.concat([train_val_1, train_val_2], axis=1)

    test_fy_1, test_fy_2 = model([data[0]['test'][0], data[1]['test'][0]])
    test_concat = tf.concat([test_fy_1, test_fy_2], axis=1)

    train_val_labels = tf.concat([train_labels, validation_labels], axis=0)

    return compute_termination_score(train_val_concat, train_val_labels, test_concat, test_labels)


def test_raw_single(train_data, train_labels, validation_data, validation_labels, test_data, test_labels):
    view_1 = tf.concat([train_data[0].T, validation_data[0].T], axis=0)
    view_2 = tf.concat([train_data[1].T, validation_data[1].T], axis=0)

    labels_1 = tf.concat([train_labels[0], validation_labels[0]], axis=0)
    labels_2 = tf.concat([train_labels[1], validation_labels[1]], axis=0)

    score_view1 = compute_termination_score(view_1, labels_1, tf.convert_to_tensor(test_data[0].T), test_labels[0])
    score_view2 = compute_termination_score(view_2, labels_2, tf.convert_to_tensor(test_data[1].T), test_labels[1])

    return [score_view1, score_view2]


@ignore_warnings(category=ConvergenceWarning)
def test_raw(train_data, train_labels, validation_data, validation_labels, test_data, test_labels):

    print(train_data[0].shape, validation_data[0].shape, test_data[0].shape)

    view_1 = tf.concat([train_data[0], validation_data[0]], axis=1)
    view_2 = tf.concat([train_data[1], validation_data[1]], axis=1)
    print('###')
    print(tf.shape(view_1), tf.shape(view_2))
    print('###')
    view_concat = tf.concat([view_1, view_2], axis=0)

    train_labels = tf.concat([train_labels, validation_labels], axis=0)

    test_concat = tf.concat([test_data[0], test_data[1]], axis=0)

    print(tf.shape(test_concat), tf.shape(view_concat), tf.shape(train_labels), tf.shape(test_labels))

    return compute_termination_score(tf.transpose(view_concat), train_labels, tf.transpose(test_concat), test_labels)