These are chat archives for nightscout/intend-to-bolus

4th
Feb 2015
diabeticgonewild
@diabeticgonewild
Feb 04 2015 02:09
Yeah, and I totally just had to contact MathWorks (MATLAB) customer support, over an error that I shouldn't be getting, due to having a license for that toolbox. I am certain that I shouldn't be getting that error, too. MATLAB is pretty solid for things.
Ben West
@bewest
Feb 04 2015 03:38
@scottleibrand re the wip/iob-cob instructions
I wonder if it'd be worth focusing on creating UI/auth to handle the setup rather than creating instructions for bunch of manual work
once the auth/UI is ready, this would essentially be tested and also ready for mainline and wouldn't need complicated setup
Scott Leibrand
@scottleibrand
Feb 04 2015 03:47
Totally agreed that should be done, but my focus right now is on closed loop stuff. Plus, you've seen my #DIYPS UI skills, right? :-p
Ben West
@bewest
Feb 04 2015 06:16
speaking of which, I've finally understood some of lane's advice wrt to something called "STPA" to connect FDA to us
would be good to get it applied to openaps, it's a way to formally argue about whether or not something is safe/controlled
diabeticgonewild
@diabeticgonewild
Feb 04 2015 06:18
Yeah, I am focused on closed loop stuff too, for right now. I got overwhelmed and I thought that data was missing from the paper I was cross-referencing, that I would need to solve for the most updated Hovorka math models (which allow you to make a type 1 into a "virtual patient" for simulation (provided a controller used in MATLAB) which is a nice safety feature such as before "closing the loop"--to catch errors). Anyways, it helps to actually read things. I "discovered" (by reading) that the missing parameters were solvable by simple division and multiplication by other parameters.
Anyways, I am tired and I made peace with some of my project, so maybe I should go to bed before I get carried away...
diabeticgonewild
@diabeticgonewild
Feb 04 2015 07:22
So, I have been calculating the values (basic multiplication/division) for the previous parameters used the older experiments for finding this one prior distribution for Hovorka's newest model. The numbers are all over the place (units are not the same, but it converts back and forth by factors of 10s, so very easy to deal with), presumably due to a bad set of samples? (which I kind of doubt). However, they do give the accuracy of each parameter estimate for each subject and parameter, in terms of a fractional standard deviation....
Once I am done "calculating" what I "need to solve for" in this paper, I guess I might have to do a propagation of error or something, due to the multiplication and division of things. IDK, I might have to turn to Google.
The accuracies are all over the place though...it makes me concerned.
diabeticgonewild
@diabeticgonewild
Feb 04 2015 07:27
like one error is 267% so that makes me a little bit relieved, sadly
diabeticgonewild
@diabeticgonewild
Feb 04 2015 08:32
Also, I am just going to wing the values, and hope they are right...I will flag the area, in case future calculations don't come out as desired.
And I am staying up all night...to the person who got me hooked on redoing my old project...you know who you are...I will get this done sooner than later.
Toby Canning
@TC2013
Feb 04 2015 14:42
I really hope you are successful Monica because I think people will be able to use your system in some neat ways. For example, I speculate that part of the reason Dana's system is so stellar is that she was and is incredibly well-controlled. I've been trying to model our system after Dana's because I think it is very safe and mimics what we are already doing, but with computer-precision. However, our child has these "seemingly" random highs all over the place (he is growing rapidly, like 4 inches this past few months). Maybe Dana's system can account for these events, but I don't think ours will. I would love to have your system running in the background where it can possibly make better predictions to handle those events. That's because those events probably aren't random and my hope is your system can make sense of the data. So, even though it would not control the pump, it would give us better data to make more informed decisions in numerous areas.
diabeticgonewild
@diabeticgonewild
Feb 04 2015 20:11
4 inches in the past few months! That's amazing!
Yeah, I hope I can get this done soon. I am able to make timelines now, for when I can finish things up, so that's a good thing.