NeuroDataDesign / orange-panda-f16s17

Automated EEG data analysis.
5 stars 5 forks source link

For April 3rd #58

Closed nkumarcc closed 7 years ago

nkumarcc commented 7 years ago

Monday Commitments

Status Task Deliverable
Present the joint prob. and kurtosis electrodes in a more clear manner @nkumarcc
Benchmark the EEGLAB functions vs ours @nkumarcc
:panda_face: RPCA Notebook with new technique, visualize changes on our data @rmarren1 note
:panda_face: EOG Regression methods @rmarren1 note
:panda_face: Comparison of pipeline versus Nicolas' preprocessed data @rmarren1 note

Thursday Checkpoint

Status Task Deliverable
:hourglass: Finish replicating EEGLAB function in Python (all of them) kurtosis @nkumarcc
:-1: Benchmark EEGLAB functions vs ours @nkumarcc
Implement and test full CMI pipeline using Matlab engine, discriminibility plots @rmarren1

Pipeline differences: certain electrodes removed Cz not in raw, in preprocessed To compare we take the intersection of the electrode channels the preprocessed and raw have (= 110 channels).

Initially, the mean squared difference between raw and preprocessed is 4.635 (frob norm of raw - preprocessed normalized by number of elements in matrix). raw

After high pass filter at .1, we get 4.631 mean squared distance, with an image like: hpf

After bandstop filter at 59 - 61 (powerline) we get 4.5 mean squared distance with an image like: bsf

Robust PCA low rank decomposition. Now we have a 1.9 mean squared distance with the following image: rpca