SSSM core example

pip install sssm --upgrade

import numpy as np
from sssm import ssm
import mne
import matplotlib.pyplot as plt

load data

raw = mne.io.read_raw_cnt(r"C:\Users\bkxcy\workspace\eeg_data_process\data\health\sleep\baoxiaoyu20211201s.cnt")
raw.load_data()

pre-process

raw = raw.filter(0.7,45)
data = (raw.resample(100).get_data(units="uV")).astype(np.float16)

Reading 0 ... 5604509  =      0.000 ... 22418.036 secs...
Filtering raw data in 1 contiguous segment
Setting up band-pass filter from 0.7 - 45 Hz

FIR filter parameters
---------------------
Designing a one-pass, zero-phase, non-causal bandpass filter:
- Windowed time-domain design (firwin) method
- Hamming window with 0.0194 passband ripple and 53 dB stopband attenuation
- Lower passband edge: 0.70
- Lower transition bandwidth: 0.70 Hz (-6 dB cutoff frequency: 0.35 Hz)
- Upper passband edge: 45.00 Hz
- Upper transition bandwidth: 11.25 Hz (-6 dB cutoff frequency: 50.62 Hz)
- Filter length: 1179 samples (4.716 s)

[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.
[Parallel(n_jobs=1)]: Done   1 out of   1 | elapsed:    0.0s remaining:    0.0s
[Parallel(n_jobs=1)]: Done   2 out of   2 | elapsed:    0.1s remaining:    0.0s
[Parallel(n_jobs=1)]: Done   3 out of   3 | elapsed:    0.3s remaining:    0.0s
[Parallel(n_jobs=1)]: Done   4 out of   4 | elapsed:    0.4s remaining:    0.0s
[Parallel(n_jobs=1)]: Done  36 out of  36 | elapsed:    4.7s finished

predict plot

model = ssm.SSM()
ret = model.predict(data[17:18],step=300)
model.plot_predictions()

input 1 epoch(30s) data

# model = ssm.SSM(model_name='ckp_last.pt')
model = ssm.SSM()
epoch_id = 200
input_data = data[17:18,epoch_id*3000:(epoch_id+1)*3000]
plt.figure(figsize=(50,4))
plt.plot(input_data.T)
plt.xlim([0,3000])
plt.show()
ret = model.predict(input_data,step=1)
model.plot_predictions()

model = ssm.SSM(model_name='ckp_last.pt')
# model = ssm.SSM()
epoch_id = 200
input_data = data[17:18,epoch_id*3000:(epoch_id+1)*3000]
plt.figure(figsize=(50,4))
plt.plot(input_data.T)
plt.xlim([0,3000])
plt.show()
ret = model.predict(input_data,step=1)
model.plot_predictions()

input full night data

model = ssm.SSM(model_name='ckp_last.pt')
ret = model.predict(data[17:18],step=300)
model.plot_predictions()

get results

df = model.to_pandas(event_threshold={
                'Spindle': 0.8,
                'Background': 0.8,
                'Arousal': 0.8,
                'K-complex': 0.8,
                'Slow wave': 0.8,
                'Vertex Sharp': 0.8,
                'Sawtooth': 0.8})
df

	Start	End	Duration	label	predict_proba	epoch_id
1	600	900	300	Arousal	0.996692	0
2	900	1800	900	Background	0.982475	0
4	2100	3300	1200	Background	0.956050	0
6	3600	4500	900	Arousal	0.982866	0
7	4500	4800	300	Background	0.995384	0
...	...	...	...	...	...	...
3296	2234100	2234400	300	Arousal	0.998787	0
3297	2234400	2235900	1500	Background	0.955171	0
3298	2235900	2240700	4800	Arousal	0.974922	0
3299	2240700	2241300	600	Background	0.932963	0
3300	2241300	2241804	504	Arousal	0.951562	0

2266 rows × 6 columns

spindles_df = df.loc[df.label=='Spindle'].reset_index(drop=True)
kc_df = df.loc[df.label=='K-complex'].reset_index(drop=True)
spindles_df = spindles_df.sort_values('predict_proba',ascending=False).reset_index(drop=True)
kc_df = kc_df.sort_values('predict_proba',ascending=False).reset_index(drop=True)

spindles_df

	Start	End	Duration	label	predict_proba	epoch_id
0	1168800	1169100	300	Spindle	0.999875	0
1	1933200	1933500	300	Spindle	0.999850	0
2	1992000	1992300	300	Spindle	0.999827	0
3	1007400	1007700	300	Spindle	0.999812	0
4	1062600	1062900	300	Spindle	0.999789	0
...	...	...	...	...	...	...
248	1856400	1856700	300	Spindle	0.802168	0
249	1246200	1246500	300	Spindle	0.802033	0
250	1549200	1549500	300	Spindle	0.801724	0
251	860700	861000	300	Spindle	0.801277	0
252	1446000	1446600	600	Spindle	0.801162	0

253 rows × 6 columns

kc_df

	Start	End	Duration	label	predict_proba	epoch_id
0	891600	891900	300	K-complex	0.941446	0
1	695700	696000	300	K-complex	0.936204	0
2	1420200	1420500	300	K-complex	0.933121	0
3	957900	958200	300	K-complex	0.923686	0
4	1834500	1834800	300	K-complex	0.915197	0
5	797400	797700	300	K-complex	0.905449	0
6	814200	814500	300	K-complex	0.887404	0
7	1919400	1919700	300	K-complex	0.886306	0
8	735900	736200	300	K-complex	0.885708	0
9	1841400	1841700	300	K-complex	0.882073	0
10	665400	665700	300	K-complex	0.880695	0
11	719400	719700	300	K-complex	0.880499	0
12	1851300	1851600	300	K-complex	0.878774	0
13	1845000	1845300	300	K-complex	0.877886	0
14	684600	684900	300	K-complex	0.873030	0
15	1908300	1908600	300	K-complex	0.871325	0
16	963900	964200	300	K-complex	0.871283	0
17	1458900	1459200	300	K-complex	0.871187	0
18	644400	644700	300	K-complex	0.865622	0
19	786300	786600	300	K-complex	0.856225	0
20	1802100	1802400	300	K-complex	0.854385	0
21	874800	875100	300	K-complex	0.853656	0
22	735300	735600	300	K-complex	0.843922	0
23	783900	784200	300	K-complex	0.842278	0
24	1133700	1134000	300	K-complex	0.836957	0
25	1793100	1793400	300	K-complex	0.836822	0
26	671700	672000	300	K-complex	0.829499	0
27	669000	669300	300	K-complex	0.826817	0
28	1126800	1127100	300	K-complex	0.825165	0
29	842100	842400	300	K-complex	0.824742	0
30	687600	687900	300	K-complex	0.822328	0
31	1243200	1243500	300	K-complex	0.819598	0
32	1305300	1305600	300	K-complex	0.818472	0
33	816000	816300	300	K-complex	0.804663	0
34	633000	633300	300	K-complex	0.802887	0

get predicted data

_id = 0
plt.plot(data[17,spindles_df.loc[_id].Start:spindles_df.loc[_id].End])

[<matplotlib.lines.Line2D at 0x18ce10fbb80>]

_id = 0
plt.plot(data[17,kc_df.loc[_id].Start:kc_df.loc[_id].End])

[<matplotlib.lines.Line2D at 0x18ce1462760>]

get feature

feature = model.feature.cpu().numpy()

feature.shape # n_epoch, n_sample, m,n

(1, 7472, 128, 15)

feature = feature[0].reshape(feature.shape[1],-1)
feature.shape# n_sample, n_feature

(7472, 1920)

# from sklearn.decomposition import PCA
from sklearn.manifold import TSNE as PCA
import pandas as pd
import seaborn as sns
pca = PCA(n_components=2, random_state=0)

transformed_feature = pca.fit_transform(X=feature)

C:\Users\bkxcy\anaconda3\envs\torch_cuda\lib\site-packages\sklearn\manifold\_t_sne.py:795: FutureWarning: The default initialization in TSNE will change from 'random' to 'pca' in 1.2.
  warnings.warn(
C:\Users\bkxcy\anaconda3\envs\torch_cuda\lib\site-packages\sklearn\manifold\_t_sne.py:805: FutureWarning: The default learning rate in TSNE will change from 200.0 to 'auto' in 1.2.
  warnings.warn(

x_component, y_component = transformed_feature[:, 0], transformed_feature[:, 1]
feature_df = pd.DataFrame({
    'x':transformed_feature[:, 0],
    'y':transformed_feature[:, 1],
    'Event Type':[['SS','KC','SW','SAW','VSW','BG','MA'][i] for i in ret],
})

# Plot our dimensionality-reduced (via PCA) dataset.
plt.figure(figsize=(10, 10), dpi=100)
# plt.scatter(x=x_component, y=y_component, c=test['labels'], cmap='tab20', s=5, alpha=8/10)
sns.scatterplot(data=feature_df,x='x',y='y',hue='Event Type',palette='bright',s=10,alpha=0.8)
plt.title('Sleep evtets classe map')
plt.legend(markerscale=1)
# plt.legend(['Spindle','K-complex','Slow-wave','Sawtooth','Vertex Sharp','Background','Arousal'])
# plt.legend()
plt.show()