DL_ECG_Classification/eval_fusion.py at main · HemaxiN/DL_ECG_Classification · 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
import torch
import numpy as np

from torch.utils.data import DataLoader
import gru as gru
import AlexNet as alexnet
import late_fusion as late
import early_fusion as early
import joint_fusion as joint

from count_parameters import count_parameters

type = ['late', 'early', 'joint'][0]

if type == "joint":
    path_weights = "save_models/joint_model_2023-02-03_06-54-57_lr0.01_optadam_dr0.3_eps200_hs256_bs1024_l20.0001"
    thresholds = [0.4761, 0.656, 0.7849, 0.8079]
    batch_size = 1024
    hidden_size = 256
    dropout = 0

elif type == "early":
    path_weights = "save_models/early_model_2023-01-26_01-04-33_lr0.001_optadam_dr0.0_eps200_hs256_bs128_l20"
    thresholds = [0.5659, 0.3952, 0.6742, 0.769]

    batch_size = 128
    hidden_size = 256
    dropout = 0

elif type == "late":
    path_weights = "save_models/late_model_2023-01-22_04-30-10_lr0.1_optadam_dr0.3_eps200_hs512_bs512_l20"
    thresholds = [0.3931, 0.645, 0.6972, 0.8535]

    batch_size = 512
    hidden_size = 512
    dropout = 0

sig_path = "save_models/grubi_dropout05_lr0005_model5"
img_path = "save_models/alexnet"

gpu_id = 0

sig_data = "Dataset/data_for_rnn/"
sig_model = gru.RNN(3, 128, 2, 4, 0.5, gpu_id=gpu_id,
                    bidirectional=True).to(gpu_id)

img_data = "Dataset/Images/"
img_model = alexnet.AlexNet(4).to(gpu_id)

if type == 'late':
    sig_model.load_state_dict(torch.load(sig_path, map_location=torch.device(gpu_id)))
    img_model.load_state_dict(torch.load(img_path, map_location=torch.device(gpu_id)))
    sig_model.eval()
    img_model.eval()

    test_dataset = late.LateFusionDataset(sig_data, img_data, sig_model, img_model, 'gru', 'alexnet',
                                          [17111, 2156, 2163], gpu_id, batch_size, part='test')

    model = late.LateFusionNet(4, 8, hidden_size, dropout).to(gpu_id)

elif type == 'early':
    sig_model.load_state_dict(torch.load(sig_path, map_location=torch.device(gpu_id)))
    sig_model.requires_grad_(False)
    img_model.load_state_dict(torch.load(img_path, map_location=torch.device(gpu_id)))
    img_model.requires_grad_(False)
    sig_model.eval()
    img_model.eval()

    img_hook = 'conv2d_5'
    sig_hook = 'rnn'
    img_model.conv2d_5.register_forward_hook(early.get_activation(img_hook))
    sig_model.rnn.register_forward_hook(early.get_activation(sig_hook))

    test_dataset = early.FusionDataset(sig_data, img_data, [17111, 2156, 2163], part='test')

    model = early.EarlyFusionNet(4, 256, 4096, hidden_size, dropout,
                                 sig_model, img_model, sig_hook, img_hook).to(gpu_id)

else:  # joint fusion
    sig_model.load_state_dict(torch.load(sig_path, map_location=torch.device(gpu_id)))
    img_model.load_state_dict(torch.load(img_path, map_location=torch.device(gpu_id)))
    sig_model.fc = joint.Identity()
    img_model.linear_3 = joint.Identity()

    test_dataset = early.FusionDataset(sig_data, img_data, [17111, 2156, 2163], part='test')

    model = joint.JointFusionNet(4, 256, 2048, hidden_size, dropout, sig_model, img_model).to(gpu_id)

test_dataloader = DataLoader(test_dataset, batch_size=1, shuffle=False)

model.load_state_dict(torch.load(path_weights, map_location=torch.device(gpu_id)))
model = model.to(gpu_id)
model.eval()

# evaluate the performance of the model
if type == 'late':
    matrix = gru.evaluate(model, test_dataloader, thresholds, gpu_id=gpu_id)
    aurocs = gru.auroc(model, test_dataloader, gpu_id=gpu_id)
else:
    matrix = early.fusion_evaluate(model, test_dataloader, thresholds, gpu_id=gpu_id)
    aurocs = early.fusion_auroc(model, test_dataloader, gpu_id=gpu_id)

count_parameters(model)

MI_sensi = matrix[0, 0] / (matrix[0, 0] + matrix[0, 1])
MI_spec = matrix[0, 3] / (matrix[0, 3] + matrix[0, 2])
MI_acc = (matrix[0, 0] + matrix[0, 3]) / np.sum(matrix[0])
MI_prec = matrix[0, 0] / (matrix[0, 0] + matrix[0, 2])
MI_f1 = (2 * matrix[0, 0]) / (2 * matrix[0, 0] + matrix[0, 2] + matrix[0, 1])
MI_auroc = aurocs[0].item()


STTC_sensi = matrix[1, 0] / (matrix[1, 0] + matrix[1, 1])
STTC_spec = matrix[1, 3] / (matrix[1, 3] + matrix[1, 2])
STTC_acc = (matrix[1, 0] + matrix[1, 3]) / np.sum(matrix[1])
STTC_prec = matrix[1, 0] / (matrix[1, 0] + matrix[1, 2])
STTC_f1 = (2 * matrix[1, 0]) / (2 * matrix[1, 0] + matrix[1, 2] + matrix[1, 1])
STTC_auroc = aurocs[1].item()

CD_sensi = matrix[2, 0] / (matrix[2, 0] + matrix[2, 1])
CD_spec = matrix[2, 3] / (matrix[2, 3] + matrix[2, 2])
CD_acc = (matrix[2, 0] + matrix[2, 3]) / np.sum(matrix[2])
CD_prec = matrix[2, 0] / (matrix[2, 0] + matrix[2, 2])
CD_f1 = (2 * matrix[2, 0]) / (2 * matrix[2, 0] + matrix[2, 2] + matrix[2, 1])
CD_auroc = aurocs[2].item()

HYP_sensi = matrix[3, 0] / (matrix[3, 0] + matrix[3, 1])
HYP_spec = matrix[3, 3] / (matrix[3, 3] + matrix[3, 2])
HYP_acc = (matrix[3, 0] + matrix[3, 3]) / np.sum(matrix[3])
HYP_prec = matrix[3, 0] / (matrix[3, 0] + matrix[3, 2])
HYP_f1 = (2 * matrix[3, 0]) / (2 * matrix[3, 0] + matrix[3, 2] + matrix[3, 1])
HYP_auroc = aurocs[3].item()

# compute mean sensitivity and specificity:
mean_mat = np.mean(matrix, axis=0)
mean_sensi = mean_mat[0] / (mean_mat[0] + mean_mat[1])
mean_spec = mean_mat[3] / (mean_mat[3] + mean_mat[2])
mean_acc = (mean_mat[0] + mean_mat[3]) / np.sum(mean_mat)
mean_prec = mean_mat[0] / (mean_mat[0] + mean_mat[2])
mean_f1 = (2 * mean_mat[0]) / (2 * mean_mat[0] + mean_mat[2] + mean_mat[1])
mean_auroc = aurocs.mean().item()
mean_g = np.sqrt(mean_spec * mean_sensi)

print('Final Test Results: \n ' + str(matrix) + '\n\n' +
      'MI: \n\tsensitivity - ' + str(MI_sensi) + '\n\tspecificity - ' + str(MI_spec) + '\n\tprecision - ' + str(MI_prec) + '\n\taccuracy - ' + str(MI_acc) + '\n\tF1 Score - ' + str(MI_f1) + '\n\tAUROC - ' + str(MI_auroc) + '\n' +
      'STTC: \n\tsensitivity - ' + str(STTC_sensi) + '\n\tspecificity - ' + str(STTC_spec) + '\n\tprecision - ' + str(STTC_prec) + '\n\taccuracy - ' + str(STTC_acc) + '\n\tF1 Score - ' + str(STTC_f1) + '\n\tAUROC - ' + str(STTC_auroc) + '\n' +
      'CD: \n\tsensitivity - ' + str(CD_sensi) + '\n\tspecificity - ' + str(CD_spec) + '\n\tprecision - ' + str(CD_prec) + '\n\taccuracy - ' + str(CD_acc) + '\n\tF1 Score - ' + str(CD_f1) + '\n\tAUROC - ' + str(CD_auroc) + '\n' +
      'HYP: \n\tsensitivity - ' + str(HYP_sensi) + '\n\tspecificity - ' + str(HYP_spec) + '\n\tprecision - ' + str(HYP_prec) + '\n\taccuracy - ' + str(HYP_acc) + '\n\tF1 Score - ' + str(HYP_f1) + '\n\tAUROC - ' + str(HYP_auroc) + '\n' +
      'mean: \n\tG-Mean - ' + str(mean_g) + '\n\tsensitivity - ' + str(mean_sensi) + '\n\tspecificity - ' + str(mean_spec) + '\n\tprecision - ' + str(mean_prec) + '\n\taccuracy - ' + str(mean_acc) + '\n\tF1 Score - ' + str(mean_f1) + '\n\tAUROC - ' + str(mean_auroc))