Feature Extraction Framework

In this notebook, we explain how to extract features segment-wise and window-wise (check the paper for theoretical details). We would also explain the possible parameters to tune the feature extraction process. PhyAAt library has function to extract Rhythemic features from EEG; either segment-wise or window-wise, so we focus on only EEG feature extraction. Custom features from EEG, PPG or GSR can be done in similar fashion. Figure below shows Feature-Extraction Framework described in paper.

Execute with

Table of Contents

Import Libraries

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from sklearn import svm

import phyaat
print('Version :' ,phyaat.__version__)
import phyaat as ph

PhyAAt Processing lib Loaded...
Version : 0.0.2

Read the data of subject=1 

dirPath = ph.download_data(baseDir='../PhyAAt_Data', subject=1,verbose=0,overwrite=False)

baseDir='../PhyAAt_Data'

SubID = ph.ReadFilesPath(baseDir)

Subj = ph.Subject(SubID[1])

Total Subjects :  1

Filtering

With Custum frequency range

Subj.filter_EEG(band =[0.5],btype='highpass',order=5)

Artifact removal using ICA

Tune threshold, windowsize, ICA method

KurThr = 2
Corr   = 0.8
ICAMed = 'extended-infomax' #picard, fastICA
winsize=128*10 # 20sec

Subj.correct(method='ICA',winsize=winsize,hopesize=None,Corr=Corr,KurThr=KurThr,
            ICAMed=ICAMed,verbose=1,
             window=['hamming',True],winMeth='custom')

ICA Artifact Removal : extended-infomax
100%|###########################################################|

Feature Extraction for T4: LWR Classification

Segment-wise feature extraction

For extracting segment-wise features set winsize=-1, which will extract 84 (=6x14) features from each segment

X_train,y_train,X_test, y_test = Subj.getXy_eeg(task=4,features='rhythmic',verbose=1,
                                  redo=False, winsize=-1,hopesize=None)


100%|##################################################|100\100|Sg - 0
Done..
100%|##################################################|100\100|Sg - 1
Done..
100%|##################################################|100\100|Sg - 2
Done..
100%|##################################################|43\43|Sg - 0
Done..
100%|##################################################|43\43|Sg - 1
Done..
100%|##################################################|43\43|Sg - 2
Done..

Shape of X_train,y_train,X_test, y_test

print('DataShape: ',X_train.shape,y_train.shape,X_test.shape, y_test.shape)
print('\nClass labels :',np.unique(y_train))

DataShape:  (290, 84) (290,) (120, 84) (120,)

Class labels : [0 1 2]

# Normalization - SVM works well with normalized features
means = X_train.mean(0)
std   = X_train.std(0)
X_train = (X_train-means)/std
X_test  = (X_test-means)/std


# Training
clf = svm.SVC(kernel='rbf', C=1,gamma='auto')
clf.fit(X_train,y_train)

# Predition
ytp = clf.predict(X_train)
ysp = clf.predict(X_test)

# Evaluation
print('Training Accuracy:',np.mean(y_train==ytp))
print('Testing  Accuracy:',np.mean(y_test==ysp))

Training Accuracy: 0.9448275862068966
Testing  Accuracy: 0.85

Window-wise Feature Extraction (2sec window)

For extracting Window-wise features with window size 2sec and 50% overlapping set winsize=128*2, which will extract 84 (=6x14) features from each window

X_train,y_train,X_test, y_test = Subj.getXy_eeg(task=4,features='rhythmic',
                                  verbose=1, redo=True, winsize=128*2)

100%|##################################################|100\100|Sg - 0
Done..
100%|##################################################|100\100|Sg - 1
Done..
100%|##################################################|100\100|Sg - 2
Done..
100%|##################################################|43\43|Sg - 0
Done..
100%|##################################################|43\43|Sg - 1
Done..
100%|##################################################|43\43|Sg - 2
Done..

Notice the number of exmaples (Shape of X, y)

Shape of X_train,y_train,X_test, y_test

print('DataShape: ',X_train.shape,y_train.shape,X_test.shape, y_test.shape)
print('\nClass labels :',np.unique(y_train))

DataShape:  (1382, 84) (1382,) (565, 84) (565,)

Class labels : [0 1 2]

# Normalization - SVM works well with normalized features
means = X_train.mean(0)
std   = X_train.std(0)
X_train = (X_train-means)/std
X_test  = (X_test-means)/std


# Training
clf = svm.SVC(kernel='rbf', C=1,gamma='auto')
clf.fit(X_train,y_train)

# Predition
ytp = clf.predict(X_train)
ysp = clf.predict(X_test)

# Evaluation
print('Training Accuracy:',np.mean(y_train==ytp))
print('Testing  Accuracy:',np.mean(y_test==ysp))

Training Accuracy: 0.9261939218523878
Testing  Accuracy: 0.8407079646017699

Tuning Feature Extraction process 

# Subj = ph.Subject(SubID[1])
# Subj.filter_EEG(band =[0.5],btype='highpass',order=5)
# KurThr = 2
# Corr   = 0.8
# ICAMed = 'extended-infomax' #picard, fastICA
# winsize=128*10 # 20sec

# Subj.correct(method='ICA',winsize=winsize,hopesize=None,Corr=Corr,KurThr=KurThr,ICAMed=ICAMed,verbose=1)

#check help to see the options
help(ph.Subject.getXy_eeg)


X_train,y_train,X_test, y_test = Subj.getXy_eeg(task=4,features='rhythmic',
                                    verbose=1,
                                    redo=False,
                                    split='serial', splitAt=100,
                                    normalize=False,log10p1=True,flat=True,
                                    filter_order=5,method='welch',
                                    window='hann', scaling='density',
                                    detrend='constant',period_average='mean',
                                    winsize=-1,hopesize=None)

Execute with