Goals for 11/2

[ewalke31]

@gkiar @ebridge2

• [x] implement slice timing correction. Deliverable: can't do mnr plots because cortex is down :( show that it "works" ie, doesn't produce BS timeseries. See html page for details. timeseries
• [x] make QC page a HTML. populate with information about what to look for for researchers. Deliverable: example of an html page. template actual
• [x] quality control improvements: add opacity as variable based on pixel intensity (probs gonna need to find some hacky way to do this, as matplotlib does not natively support it). Make the histograms less useless. Deliverable: see html page for details.
• [x] restful calls to get calcium imaging data, and then check QC on calcium imaging data. Deliverable: QC figures. PIVOT: can't find documentation on pulling timeseries data, and no responses on my issue, so pivotted to run QC on aligned DTI data instead. figure

deliverable by next week:

• [x] PCA/CompCor overview; outline of how it will fit into FNGS, why, potential benefits (texfile). Note that this will tie into ML final project (deliverable with Tanay). overview
• [ ] docker that runs FNGS.

@ewalke31

• [x] implement EM Hidden Markov Model (HMM) with Tanay (deliv: code.md pseudo & sims)
  http://neurodatadesign.github.io/fngs/ewok_weekly/EM_HMM.html
• [ ] future goal (if time): implement EM KFS (deliv: code.md pseudo & sims)

@02agarwalt

• [x] EM for Hidden Markov Model: code.md with ewok

by wednesday:

• [x] PCA/CompCor overview; outline of how it will fit into FNGS, why, potential benefits (markdown file). Note that this will tie into ML final project (deliverable with Eric).

[ebridge2]

Todos for morning:

• [x] add figure showing qc works with different modality. see above for link.

[tsunwong123]

@ebridge2 Great job on the HTML template. I think the PCA/CompCor for fMRI for your final project sounds like a great idea as well.

[rmarren1]

@ebridge2 doing a similar thing for representation learning final project, professor suggested FPCA for high dimensional time series data. i'm probably going to do a code.md type thing to understand it so i'll link you

[jovo]

@ebridge2

• i can't see the html page?
• mention me & @perlman in your issue re:freeman14
• let's get that web-service live, take @gkiar out of critical path?

@ewalke31

• probabilities translate into R code via matrices whose columns sum to one.
• rabiner 1989 is my favorite review on the subject
• do you think you understand EM for HMM/KFS now? perhaps try using a package now?
• perhaps give it longer history for "poor performance" setting, write out the MLE and see which approximations are made to the joint distribution
• shall we meet to discuss HMM?

[nkumarcc]

@ewalke31 @02agarwalt try doing a more conceptual focus for a week instead of thinking about doing anything coding wise just write a very comprehensive LaTeX file explaining equations (mine for KDE is here) I'm guessing this is the case for you guys, but KDE had a lot of very esoteric papers and lectures that I had to read in order to get out the basics that really only took up a page but was the real core of it. (Also I know yours looks more complicated than mine, but maybe a good idea if you're having trouble with understanding it)

[ebridge2]

@jovo

1. Didn't get a chance to update the issue before class, as greg and I were talking prior and I made a fix (used to be absolute indexed images, hence I just linked a template since it wouldn't work unless you had the same directory organization as the user, but changed to relative indexing at greg's advising) If you go here is an example with relative paths.
2. Done
3. Sounds good; I'll make the docker a priority for this week and then web service the next (need to make some progress on the tGMM stuff, so can't commit fully to having the entire web service up in one week).

[mlee156]

@ewalke31 @02agarwalt Great work, don't get discouraged because something's not working or has bugs - my histogram equalization took way longer to implement then it should have but because of the experience I learned a lot about computer vision since I had to research opencv so much!

[mnatenz1]

@ewalke31 If you could write a brief description above your pseudocode to describe what the EM Hidden Markov Mode does, it would definitely make it easier for readers understand what's going on, and why you're doing it.