For April 16th

[rmarren1]

Monday Commitments

Status	Task	Deliverable
:panda_face:	Docker Image which will run our pipeline on one individual, for use in batch. Use new better plots.	dockerhub github results on neurodatadesign-test bucket. @rmarren1
:panda_face:	Test interpolation method to replace SSI, new discriminibility.	see below @rmarren1
:-1: :	Full pipeline AWS integrated	@nkumarcc
:panda_face: :	Get a "hello world" AWS Batch instance running such that our pipeline can easily be added to it	dir with files, need to run locally @nkumarcc
:panda_face: :	Updated flow chart	cloudy flowchart diagram @nkumarcc

Discriminibility results from most recent test:

Data setting:

• 4 subjects (4 randomly selected from Nicolas' data)
• 11 trials per subject (each trial is one CxT data matrix from a paradigm, C = 128, T is 70,000 to 200,000).
• Derivative on which metric is computed: Either correlation matrix, or thresholded correlation matrix with identity subtracted (graph with no self loops)
• Metrics: Frob. norm (of correlation matrix) or difference in number of 3/4 cycles (on thresholded correlation matrix)

Right hand side markers is the scores of Nicolas' preprocessed data for each transform/metric pair. Bottom should read: setup, highpass, bandstop, bad_detect, wavelet_coefficient_interp, eog_regress, rpca_denoise.

[rmarren1]

@nkumarcc @gkiar

[jovo]

@rmarren1 it seems that only eog_regress is useful. what is discriminability missing?