NSGCCA/solver_hsic.py at main · Rows21/NSGCCA · 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
#import torch
import numpy as np
from tqdm import tqdm
#torch.set_default_tensor_type(torch.DoubleTensor)

from Simulation.proposedmodels.synth_data import create_synthData
from Simulation.proposedmodels.hsic_sgcca import HSIC_SGCCA

from Simulation.proposedmodels.utils import rbf_kl, rbf_kx
import itertools

import time
#torch.manual_seed(0)
class Solver():
    def __init__(self, device='cpu'):
        self.SNGCCA = HSIC_SGCCA(device)
        self.device = device

    def tune_hyper(self, x_list, k=5, mode = 'cv', a=[1e-5, 1e-5, 1e-5]):
        # split
        shuffled_index = np.random.permutation(len(x_list[0]))
        split_index = int(len(x_list[0]) * 1/k)
        fold_index = []
        for j in range(k):
            start_index = j * split_index
            end_index = (j + 1) * split_index
            fold_index.append(shuffled_index[start_index:end_index])

        obj_validate = 999999
        if mode == 'cv':
            # set hyperparams set
            #a = [1e-4, 1e-4, 1e-4]
            print("Start Hyperparams Tuning")
            # fixed folds number
            # start cross validation
            b0 = a
            count = 0

            # for a in combinations_with_replacement(a, 3):
            while max(a) < 1e-1:
                a = [i * 10 for i in a]
                count +=1
                o_list = [0] * 3
                obj_temp = np.zeros((k, len(x_list)))
                min_index, mean_obj = self._cv(x_list, fold_index, shuffled_index, obj_temp, o_list, a, mode, k)
                print("Sparsity selection number=", count, "hyperparams=", a, "obj=", mean_obj)
                #a[min_index] = a[min_index] * 10

                if mean_obj < obj_validate:
                    b0 = a
                    #print(b0)
                    obj_validate = mean_obj
                else:
                    continue

            print("Finish Tuning!")
            return b0, obj_validate
        elif mode == 'multi_start':
            a = a
            b0 = None
            print("Start Multi-start Tuning")
            for i in tqdm(range(k)):
                o_list = [0] * 3
                obj_temp = np.zeros((k, len(x_list)))
                if i == 0:
                    Pi0_list = [np.eye(x_list[j].shape[1])/np.trace(np.cov(x_list[j].T)) for j in range(3)]
                Pi_list, u_list, obj = self.SNGCCA.fit_admm(x_list, constraint=a, criterion = 5e-5, logging=2, mode=mode, Pi0_list=Pi0_list)
                if obj > obj_validate:
                    b0 = self.SNGCCA.Pi0_list
                    obj_validate = obj
                else:
                    continue

            return b0

    def _cv(self, x_list, fold_index, shuffled_index, obj_temp, o_list, a, mode, k=5):
        for fold in tqdm(range(k)):

            fold_mask = np.zeros_like(x_list[0], dtype=bool)
            fold_mask[fold_index[fold]] = True

            train_data = []
            test_data = []

            for _, view in enumerate(x_list):

                test_data.append(view[fold_index[fold], :])
                non_fold_index = [num for num in shuffled_index if num not in fold_index[fold]]
                train_data.append(view[non_fold_index, :])

            Pi_list, u_list = self.SNGCCA.fit_admm(train_data, constraint=a, criterion = 1e-4, logging=2, mode=mode)
            K_list = [rbf_kx(test_data[i], Pi_list[i]) for i in range(len(test_data))]
            obj_k = []
            for items in itertools.combinations(range(len(K_list)), 2):
                obj_k.append(np.trace(K_list[items[0]] @ rbf_kl(K_list[items[1]])))
            obj_temp[fold] = np.stack(obj_k)
            o_list = [np.sum(abs(u_list[i]) < 0.05) + o_list[i] for i in range(3)]
        len_u = [len(u) for u in u_list]
        mean_obj = np.mean(np.sum(obj_temp, axis=1))
        min_index = np.argmin(np.stack(o_list)/np.stack(len_u)/k)#np.argmax([np.mean(np.abs(i)) for i in u_list])

        return min_index, mean_obj


    def _get_outputs(self, b, logging=1):
        u = self.SNGCCA.fit_admm(b, logging)
        return u


if __name__ == '__main__':
    ############
    # Hyper Params Section
    #device = 'cuda' if torch.cuda.is_available() else 'cpu'
    #print("Using", torch.cuda.device_count(), "GPUs")
    device='cpu'
    import numpy as np
    mode = 1
    N = 100
    num = 5
    tol = 100

    Pi = None

    views = create_synthData(num,N, mode=mode, F=tol)
    solver = Solver(device=device)

    constraint, _ = solver.tune_hyper(views, k=5, mode='cv')
    print(constraint)
    #Pi = solver.tune_hyper(views, k=1, mode='multi_start', a=constraint)
    criterion = 5e-3
    u = solver.SNGCCA.fit_admm(views, constraint=constraint, criterion=criterion, logging=0)
    print(u[0])
    #print(u[1])
    #print(u[2])