[Mne_analysis] Request for help regarding Selecting The Best Number Of Components For TSVD

MD KHORSHED ALAM khorshed.alam at live.iium.edu.my
Tue May 15 02:52:05 EDT 2018
Search archives:

        External Email - Use Caution        

Respected Sir,

Hope you find this email in best of your health.

My research area revolves around EEG data dimensionality reduction and aggregation for remote monitoring application. I am currently trying to reduce dimensionality of EEG data using TSVD. I want to select the number of components based on the threshold of the Explained Variances ratio. For example,

In PCA-     pca = PCA(n_components=0.995).

Here is my current code:

f = scipy.io.loadmat('D:\PhD Project\m.mat')
#print(f)
data=f['seg']
data_con = np.empty((23, 0))

for segment_index in xrange(10):        #xrange(seg_data.shape[0]):

                X = data[segment_index][0]
                #print (X)
                pca = PCA(n_components=0.99000, whiten=True, svd_solver='full')
                d=pca.fit(X)
                varince= d.explained_variance_
                #print("varience", varince)
                #plt.plot(varince)
                #show()
                #print("varince_ratio",d.explained_variance_ratio_)
                varience_sum= pca.explained_variance_ratio_.sum()
                print("varience sum", varience_sum)
                reduced_data = pca.transform(X)

                data_con = np.concatenate((data_con, reduced_data), axis=1)
                #data_con.append(reduced_data)

                print (reduced_data.shape)
                #print(reduced_data)

                s_values=pca.singular_values_
                print("singular values",s_values)
                #print("sum of singular values",pca.singular_values_.sum())
                #plt.plot(reduced_data)
                #show()

print(data_con.shape)

OUTPUT:


Automatically created module for IPython interactive environment
('varience sum', 0.99507757805869179)
(23L, 5L)
('singular values', array([ 411.06935131,  112.96920549,   62.14858432,   44.93381988,
         40.8842142 ]))
('varience sum', 0.99440716181852129)
(23L, 4L)
('singular values', array([ 487.96847864,  223.39657605,   73.57105848,   43.53144261]))
('varience sum', 0.99255786598154405)
(23L, 3L)
('singular values', array([ 667.84204849,   91.52957156,   45.24156259]))
('varience sum', 0.99432236995180889)
(23L, 2L)
('singular values', array([ 857.48396352,   93.0594852 ]))
('varience sum', 0.99071739908561018)
(23L, 1L)
('singular values', array([ 817.06268687]))
('varience sum', 0.99579132597569364)
(23L, 3L)
('singular values', array([ 671.58034853,  157.19757837,   53.00931532]))
('varience sum', 0.99924045901363834)
(23L, 2L)
('singular values', array([ 2149.55587243,   506.57406572]))
('varience sum', 0.99324682816663179)
(23L, 2L)
('singular values', array([ 915.25285164,   65.15344918]))
('varience sum', 0.99561247009259601)
(23L, 5L)
('singular values', array([ 229.48540692,  216.6580379 ,   52.97195239,   36.59666415,
         24.51371431]))
('varience sum', 0.99294116111319852)
(23L, 3L)
('singular values', array([ 423.79775695,  282.11919481,   60.35361031]))


Concatenate_metrices (23L, 30L)


I want to do same way on Truncated SVD but couldn't select the following way. For example- TruncatedSVD(n_components= 0.99)

 It will be a great help for me if you can advise me to solve my earlier mentioned issue on the best number of component selection based on explained variance ratio. As it will really help me in developing my data aggregation technique.

Looking forward to your feedback.

Thank You.

Alam
Ph.D. Researcher.






Thanks With Warm Regards,

MD. KHORSHED ALAM (Shishir)

Graduate Research Assistant (GRA)

Center of Intelligent Signal and Imaging Research (CISIR)
Department of Electrical and Electronic Engineering
Universiti Teknologi PETRONAS
Bandar Seri Iskandar
32610 Tronoh
Perak Darul Ridzuan
Malaysia

Mobile no: +60176459080
Email ID: khorshed.alam at live.iium.edu.my<mailto:khorshed.alam at live.iium.edu.my>
Alternative :shishir_lmu at yahoo.com<mailto:shishir_lmu at yahoo.com>, alam.0213 at gmail.com<mailto:alam.0213 at gmail.com>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20180515/7b997a08/attachment-0001.html 


More information about the Mne_analysis mailing list