PhysioNet/test_pairwise_dtw.py at master · hsd1503/PhysioNet · 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
from resnet1d.util import read_data_physionet_4
import numpy as np
from dtw import dtw, accelerated_dtw
import timeit
import matplotlib.pyplot as plt
from tqdm import tqdm
import wfdb
from wfdb import processing
from collections import Counter
from joblib import Parallel, delayed

def plot_longterm_dtw():
    X_train, X_test, Y_train, Y_test, pid_test = read_data_physionet_4()
    print(X_train.shape, Y_train.shape)
    np.random.seed(0)

    for warp in [1,2,3]:

        for _ in tqdm(range(50)):
            idx1 = np.random.choice(list(range(X_train.shape[0])))
            idx2 = np.random.choice(list(range(X_train.shape[0])))
            seg1 = np.squeeze(X_train[idx1])
            seg2 = np.squeeze(X_train[idx2])
            seg1 = seg1 - np.min(seg1) + 3
            seg2 = seg2 - np.max(seg2) - 3

            start = timeit.default_timer()
            d, cost_matrix, acc_cost_matrix, path = accelerated_dtw(seg1, seg2, dist='euclidean', warp=warp)
            end = timeit.default_timer()
            print ("time: {} s".format(end-start))
            plt.figure(figsize=(20,3))
            plt.plot(seg1)
            plt.plot(seg2)
            for i in range(0, len(path[0]), 50):
                x1 = path[0][i]
                x2 = path[1][i]
                y1 = seg1[x1]
                y2 = seg2[x2]
                plt.plot([x1, x2], [y1, y2], c='r')
            plt.savefig('img_dtw_rhythm/{}_{}_{}_{:.4f}.png'.format(warp, idx1, idx2, d))

def plot_shortterm_dtw():

    wrap = 1
    # average wave dtw
    X_train, X_test, Y_train, Y_test, pid_test = read_data_physionet_4()
    print(X_train.shape, Y_train.shape)
    np.random.seed(0)

    for _ in tqdm(range(300)):
        idx1 = np.random.choice(list(range(X_train.shape[0])))
        idx2 = np.random.choice(list(range(X_train.shape[0])))
        seg1 = np.squeeze(X_train[idx1])
        seg2 = np.squeeze(X_train[idx2])
        beat1 = get_avg_beat(seg1)
        beat2 = get_avg_beat(seg2)
        beat1_dist = beat1 - np.mean(beat1)
        beat2_dist = beat2 - np.mean(beat2)
        beat1_plot = beat1 - np.min(beat1) + 3
        beat2_plot = beat2 - np.max(beat2) - 3

        start = timeit.default_timer()
        d, cost_matrix, acc_cost_matrix, path = accelerated_dtw(beat1_dist, beat2_dist, dist='euclidean', warp=wrap)
        end = timeit.default_timer()
        print ("time: {} s".format(end-start))
        plt.figure(figsize=(5,3))
        plt.plot(beat1_plot)
        plt.plot(beat2_plot)
        for i in range(0, len(path[0]), 10):
            x1 = path[0][i]
            x2 = path[1][i]
            y1 = beat1_plot[x1]
            y2 = beat2_plot[x2]
            plt.plot([x1, x2], [y1, y2], c='r')
        plt.savefig('img_dtw_beat/{:.4f}_{}_{}.png'.format(d, idx1, idx2))

def get_avg_beat(seg, window=200, fs=300):
    out = []
    left_offset = window//2
    right_offset = window - left_offset
    len_seg = len(seg)
    qrs_inds = processing.xqrs_detect(sig=seg, fs=fs, verbose=False)
    if len(qrs_inds) == 0:
        return np.zeros(window)
    for ind in qrs_inds:
        if ind >= left_offset and ind <= (len_seg-right_offset):
            out.append(seg[(ind-left_offset):(ind+right_offset)])
    out = np.array(out)
    if len(out.shape) == 1:
        avg_beat = out
    else:
        avg_beat = np.mean(out, axis=0)
    return avg_beat

def get_all_beat():
    """
    this code is a piece of mess, don't look it.
    """
    window = 200
    np.random.seed(0) # this line is important
    X_train, X_test, Y_train, Y_test, pid_test = read_data_physionet_4()
    print(X_train.shape, Y_train.shape)
    print(np.sum(X_train[0])) # 173.1752265279791
    n_sample = X_train.shape[0]
    all_beat = []
    for i in tqdm(range(n_sample)):
        seg = np.squeeze(X_train[i])
        beat = get_avg_beat(seg, window=window)
        if len(beat) == 1:
            beat = np.zeros(window)
        all_beat.append(beat)
    all_beat = np.array(all_beat)
    print(all_beat.shape)
    np.save('all_beat.npy', all_beat)

    out = []
    all_beat = np.load('all_beat.npy', allow_pickle=True)
    for i in range(len(all_beat)):
        if len(all_beat[i]) == 0:
            all_beat[i] = np.zeros(window)
        else:
            out.append(list(all_beat[i]))
    all_beat = np.array(all_beat)
    print(all_beat.shape)
    all_len = [type(_) for _ in all_beat]
    print(Counter(all_len))
    out = np.array(out)
    print(out.shape)
    np.save('all_beat.npy', out)

def compute_batch(i_batch):
    start_idx = i_batch*batch_size
    end_idx = start_idx + batch_size
    if end_idx > n_sample:
        end_idx = n_sample

    batch_beat = all_beat[start_idx:end_idx]
    print(batch_beat.shape)
    local_batch_beat = batch_beat.shape[0]
    mat = np.zeros((local_batch_beat, local_batch_beat))

    for i in tqdm(range(0, local_batch_beat-1)):
        for j in range(i+1, local_batch_beat):

            beat1 = batch_beat[i]
            beat2 = batch_beat[j]
            beat1_dist = beat1 - np.mean(beat1)
            beat2_dist = beat2 - np.mean(beat2)

            d, cost_matrix, acc_cost_matrix, path = accelerated_dtw(beat1_dist, beat2_dist, dist='euclidean', warp=wrap)
            mat[i,j] = d

    np.save('sim_mat/{}.npy'.format(i_batch), mat)


if __name__ == "__main__":

    np.random.seed(0) # this line is important

    all_beat = np.load('all_beat.npy')
    n_sample = all_beat.shape[0]
    print(all_beat.shape)
    wrap = 1
    batch_size = 256
    all_batchs = list(range(n_sample//batch_size+1))
    # all_batchs = all_batchs[:10]
    print(all_batchs)

    # Parallel(n_jobs=28)(delayed(compute_batch)(i) for i in all_batchs)

    compute_batch(all_batchs[-1])