Will show, while demoing with Ryan and by myself @nkumarcc
Comments:
Need to figure out a batch instance size that doesn't cost too much $$$. We only need ~2 gigs of ram, and runs on a single core for 5 - 15 minutes depending on size of trial and depending on what derivatives we want. When choosing m3, it automatically launches a 2xlarge... how can we get that down?
Should coherence be a function of frequency? In the code we have it is, e.g. the 'coherence' between two channels is a vector rather than a single number. It seems we are using a windowed version and should just calculate over the entire channel.
When files are loaded back into s3 they are not public, need to fix
Make nice plots like Eric has for performance (RAM usage, compute time)
Wrap together all the code, create a pip installable package of the pipeline methods.
Justify every pipeline change we made. Specifically, wavelet interpolation and removing RPCA. Numerical proof ours works better (discriminibility).
Run on a larger data set and make group analysis launch after all jobs are done in a dataset.
Monday Commitments
Comments:
Thursday Commitments
Not sure a good checkpoint for this one.