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

18th
Jan 2015
Chris Oattes
@cjo20
Jan 18 2015 04:29
@scottleibrand where did you get the algorithm for insulin remaining / total activity etc from?
Chris Oattes
@cjo20
Jan 18 2015 06:15
oh and where on earth did 2/dia/60/peak come from?
Scott Leibrand
@scottleibrand
Jan 18 2015 07:41
I derived the IOB curve from a simplified insulin activity "curve" that goes linearly from zero to peak, and then back down to zero at the end of DIA.
So a big ^ shape. Peak is at 75m for a 3h DIA IIRC
Partly that was based on insulin activity curves from the literature, partly based on "what works" for us.
@kenstack has some better more complex curves based on more recent literature that he will be committing to wip/iob-cob soon.
diabeticgonewild
@diabeticgonewild
Jan 18 2015 09:33

FYI, if you are using the most updated Hovorka models, for the really scary math, you will likely have to use WinBUGS (with WBDiff and GeoBUGS), which can be used within MATLAB via an inline function using a specific MATLAB toolbox.

The really scary math is not as bad as it appears. Hovorka published some of his WinBUGS code (although it is an autoregressive model, not a Markov Chain Monte Carlo), which makes the dirty work of implementing the models a lot easier to properly do. http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3201840/

kenstack
@kenstack
Jan 18 2015 17:02
Nice reference @diabeticgonewild - question - the glucose dynamics portion looks to be quite simple then they layer on to the basic model a far more complicated regression to help predictions based on as they say both inter person and inter group variability. It's interesting they try to account for cgm directly in the regression. In your experience are the renal and hepatic terms in the original hovorka Lehman models important over say a 3-5 hr
period after a bolus ? My initial estimates say no but they always include them in the early models - then again their timeframes are longer due to the slower insulin they were modeling and their time frames for the model were days not hours
diabeticgonewild
@diabeticgonewild
Jan 18 2015 19:02
I have only tested out my old artificial pancreas model in MATLAB from 2011, using CareLink exports (and therefore had to do "illegal operations" such as sum the "correction basals" into a single bolus), so it wasn't real-time.
Hepatic modeling is important as it is a variable that is affected directly by the ingestion of food. Renal modeling is important to ensure more stability of the loop, if it has already become unstable due to the glucose dumping.
diabeticgonewild
@diabeticgonewild
Jan 18 2015 19:09
I think they do an autoregressive model with finding the residuals for the CGM values primarily as an attempt to deal with the (unknown) error, which could not be modeled at the time. Also, for the sake of creating a virtual patient population, at the time, it was a way of getting more realistic data besides using Gaussian white noise with a zero mean and unit covariance for error modeling, which is clearly wrong, as glucose levels are skewed and distrubuted in the upwards direction, whether or not the individual is healthy or not.
diabeticgonewild
@diabeticgonewild
Jan 18 2015 19:25

Actually, looking at the paper I am using for my implementation of the mathematical models, renal and hepatic models have been substituted, using time-varying piecewise linear fluxes in various models:

"The novel component of the present work is the additive time-varying piecewise-linear flux f_sub_g_of_t (μmol/kg/min) to model diurnal changes and other time-varying characteris- tics of glucose kinetics. The plasma glucose concentration G(t) (mmol/l) is obtained as..."

diabeticgonewild
@diabeticgonewild
Jan 18 2015 19:32
But, endogenous glucose production is still accounted for, but the hepatic models are ditched. Glucose disposal is also accounted for. Basal rates are changed at approximately every 15 minutes in the Cambridge trials. But keep in mind that the time-varying piecewise linear fluxes are calculated in 15 minute intervals.
Ben West
@bewest
Jan 18 2015 20:51
useful stuff for gist/blog/code
diabeticgonewild
@diabeticgonewild
Jan 18 2015 20:52
OK, will do.
Ben West
@bewest
Jan 18 2015 22:10
interesting burrito experience last night
went for dual wave, 30/70
worked great for first 5 hours or so
then for next 6 hours kept going
corrected when I woke up wondering if there might some be left still
so took wizard's recommendation with tiny bit extra
went out of house, walked around tiny bit and boom, huge plummet
heh, burrito was really done, but really interesting how quickly the glucose present evaporated
like sensitivity was uh increased
like it had maxed out at 400+ for couple hours at 6/7
Chris Oattes
@cjo20
Jan 18 2015 22:14
how many carbs was it?
Ben West
@bewest
Jan 18 2015 22:14
then in 90 minutes it dropped from 400+ to 55
so that's the thing, the initial dosing was pretty good
just needed little more basal on end of tail
fatter/longer tail on carbs than insulin
but it was the rapid drop that was interesting
usually takes bit longer
400 points in 90 min
Dana Lewis
@danamlewis
Jan 18 2015 22:15
Are you sure there was no pooled insulin?
Ben West
@bewest
Jan 18 2015 22:15
no pooled insulin?
I was walking around without breakfast
which is bit unusual, more activity than usual
but even still, before started walking it was at 90
and I was like oh, 2 more units on board, time for something delicious
Dana Lewis
@danamlewis
Jan 18 2015 22:16
If you were inactive, then active, sometimes if there's insulin in a puddle in your body, it can cause a whoosh
Ben West
@bewest
Jan 18 2015 22:16
ah
could be
was definitly a whoosh morning
boom, no more 400
the delicious thing was pretty good too
Chris Oattes
@cjo20
Jan 18 2015 22:17
heh
Dana Lewis
@danamlewis
Jan 18 2015 22:17
I have that happen in am - BGS are perfect flat overnight, no IOB, 10 min walk to work, whoosh to a low, only causation would be puddled/pooled insulin from sometime overnight from how I was sleeping
:)
Ben West
@bewest
Jan 18 2015 22:17
hmm, dunno
Chris Oattes
@cjo20
Jan 18 2015 22:17
I really wish I understood what on earth my body is doing. 10.4, did 2.35u. 90 mins later, I've got a flat line at 9.7
Ben West
@bewest
Jan 18 2015 22:17
never heard of insulin pooling, you mean physically?
Chris Oattes
@cjo20
Jan 18 2015 22:18
@bewest I guess if you lie on the pump site it reduces blood flow to the area, so insulin will just sit there
Ben West
@bewest
Jan 18 2015 22:18
or is it just increased blood flow in general
Dana Lewis
@danamlewis
Jan 18 2015 22:18
Yes. Physically around the site or some other spot for some reason
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:18
Yeah, correct @danamlewis , there is only so much insulin you can take at a certain time. There is a ratio where the insulin volume : surface area becomes substantially different, and insulin absorption becomes unreliable
Ben West
@bewest
Jan 18 2015 22:18
hmm, interesting
figured it had more to do with gastric emptying/glycogen pool/fasting
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:19
It happens around ~7 units in one area of an injection site
Dana Lewis
@danamlewis
Jan 18 2015 22:19
It's not dependent on volume in my experience, more blood flow to the area, usually sitting/laying a certain way
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:19
Gastric emptying can be highly variable, based on actual BG level.
Ben West
@bewest
Jan 18 2015 22:19
on theory of when glycogen store is empty, insulin is more "effective"
Dana Lewis
@danamlewis
Jan 18 2015 22:20
@bewest could be, but activity is a flag for me to guess the pooling issue. Can't measure or control it, but pretty strong correlation if not causation
Ben West
@bewest
Jan 18 2015 22:20
yeah
yeah, tricky to understand what's happening
more tools :-)
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:22
Yeah, I have started programming the new project, but it's really hard to get really into it, as I know I will drive myself crazy for days on end when I get hung up.
Ben West
@bewest
Jan 18 2015 22:22
`git commit` all the things :-)
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:23
?
Ben West
@bewest
Jan 18 2015 22:23
I'm :+1: 'ing your programming
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:24
So I found an awesome book that handles the math that I am uncomfortable with. I think I will have it all covered, for like implementing it, except for getting hung up on the programming. http://www.amazon.com/dp/0387715983/ref=wl_it_dp_o_pC_nS_ttl?_encoding=UTF8&colid=2OYKR43UGE0YB&coliid=IP6IS0K1ZILOE
Oh, you want my code, too?
Ben West
@bewest
Jan 18 2015 22:24
nice, make sure it's in the gist under your recommended reading section
all I really care about is the code :-)
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:25
LOL, if you say so, but that's really risky to do...I am going to do it all in MATLAB first then port it to like C
I have to think about it. I know it's extremely risky.
Ben West
@bewest
Jan 18 2015 22:26
putting code in a private repo is certainly not risky
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:26
If someone uses it, yes.
Ben West
@bewest
Jan 18 2015 22:26
neither are tracking changes to it using tools like git
that's a big if
lots of ifs involved
that said, I do worry about some of the dangerous tools I put out there
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:27
true
Ben West
@bewest
Jan 18 2015 22:27
I also make sure they have warnings and are hard enough to use that you have to prove you probably know how to use the tool in order to use it
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:28

So, if you are into implementing original models (I'm not...I will leave it up to the people at Cambridge) there's a book that's a gold mine

See page 78. They have a whole chapter for the artificial pancreas

Yeah, but I have to deal with engineering ethics
Ben West
@bewest
Jan 18 2015 22:28
I pretty much only look at code :-)
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:29
It's like DIY AP Modeling for dummies
Yeah, me too.
Ben West
@bewest
Jan 18 2015 22:29
check out nancy leveson http://sunnyday.mit.edu/, she makes an effective argument saying that code is ephemeral and essentially not dangerous
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:31

Some academic is not going to win against US law. They can make an example out of me. Plus, I wouldn't be nervous unless I didn't have rare autoimmune diseases. Like getting arrested/incarcerated/detained is not really an option for me.

When I was younger, before I knew I had rare autoimmune diseases, I may have felt differently.

If it was strictly on a one-on-one basis, and I gave the code in analog form (e.g. paper), I wouldn't be that nervous.
Chris Oattes
@cjo20
Jan 18 2015 22:32
I'm not sure what you think you'd get arrested for?
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:33
Uh, like my cousin is like an electrical engineer who works on Class II medical devices. An artificial pancreas would be classified as Class III, which goes in the bracket of ventilators and everything else. Any sort of implementation of a project, that is classified even on a personal level, is risky.
Ben West
@bewest
Jan 18 2015 22:34
correct, everything here, including nightscout is class iii
Dana Lewis
@danamlewis
Jan 18 2015 22:34
No, it's not. Only if you're doing something commercial with it.
Ben West
@bewest
Jan 18 2015 22:34
that's why I published it and am personally sponsoring them with FDA
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:35
So if I post my code, I could be OK?
Dana Lewis
@danamlewis
Jan 18 2015 22:35
Fda regulates based on commercial activity. DIY implementations and individual research, no.
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:35
It's still a liability, civillaly, I know.
Ben West
@bewest
Jan 18 2015 22:35
for someone who is simply putting up code for themselves and not recommending it to anyone, the FDA doesn't care at all
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:35
Plus engineering ethics.
Ben West
@bewest
Jan 18 2015 22:35
you are not a victim from this perspective... there's nothing stopping you from putting up code
and I say this as a person who has and is publishing very dangerous things indeed
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:36
OK, got it, but I have to test it before I go wild and post it. I really can't get in to trouble with the law....and I have a history of doing risky (not necessarily bad) things.
Ben West
@bewest
Jan 18 2015 22:36
as far as leveson, she's got a pretty sterling track record testifying to congress and other organizations, it's why they let her right the book
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:36
I have never been arrested or anything.
Ben West
@bewest
Jan 18 2015 22:36
right, I tend to see that as a bit of hubris
what would you be arrested for?
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:36
But I am just saying.
I was a pothead back in the day, but my gastroparesis was out of control, and I was prescribed marijuana pills anyways (Marinol).
So it's not as bad as it sounds.
Ben West
@bewest
Jan 18 2015 22:37
I don't see how any of it is relevant
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:37
Plus I am fat now and I need to lose weight.
Ben West
@bewest
Jan 18 2015 22:37
dunno how you think regulation/police works
but they don't have resources to be going out and arresting people for posting code no one is using and is for educational/research purposes
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:38
No, I am just saying. I have reasons to be paranoid. I really can't get arrested or put myself at risk.
Chris Oattes
@cjo20
Jan 18 2015 22:38
and what bewest is saying is that there is no risk
Ben West
@bewest
Jan 18 2015 22:38
I don't see what connection those fears have to reality
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:39
I know, law enforcement has to prioritize.
Ben West
@bewest
Jan 18 2015 22:39
it's not even illegal
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:39
OK, got it.
Chris Oattes
@cjo20
Jan 18 2015 22:39
you can't put yourself at risk, and there is no risk. It's a perfect match!
Ben West
@bewest
Jan 18 2015 22:39
why would they arrest you?
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:39
or get sued?
Ben West
@bewest
Jan 18 2015 22:40
anyone can sue you, that has nothing to do with the authorities or being arrested
as-is, you have no money
there's no point to suing you
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:40
yeah, but getting in trouble financially is really bad when you have extremely expensive medical conditions.
and I am not talking about diabetes at all.
I mean, I can hook people up. I just want to be really low key.
Ben West
@bewest
Jan 18 2015 22:42
that's true, if you think your code is so dangerous and will be so popular as to kill people you shouldn't publish it, ethically
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:43
I mean, I will probably give in, but I can't make any guarantees. I have to think about it a lot. There is so much risk and a lot of stipulations involved, personally, professionally, and ethically.
Ben West
@bewest
Jan 18 2015 22:43
dunno, I don't try not to take myself that seriously
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:44
yeah, I guess. I used to be a pretty chill person, but that is another story.
Ben West
@bewest
Jan 18 2015 22:44
there are several significant hurdles to cross to create that kind of problem
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:44
True. A lot.
Ben West
@bewest
Jan 18 2015 22:45
and some of them you would have to willingly cross
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:45
yeah, and it's hard to pull that off.
Ben West
@bewest
Jan 18 2015 22:45
so to prevent myself from doing something because of a minute hypothetical risk
these papers and things aren't going to help unless there is code implemented and maintained by dozens of people
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:46
True. It's just a hard thing to get caught up in.
I can tell you that it won't work that well it GitHub, as the code is very processor-intensive.
Scott Leibrand
@scottleibrand
Jan 18 2015 22:47
Monica: publishing (code, results, whatever) is free speech. FDA only cares if you distribute a class 3 medical device. And just code does not a medical device make: you have to provide an assembled solution, or detailed directions so that general end users can build a device themselves.
Chris Oattes
@cjo20
Jan 18 2015 22:47
I must remember to take my mongodb username/password out of my code before I put it on git
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:47
true.
Publishing to GitHub I guess is a starting point. However, I am not publishing it without doing at least some basic tests as I don't want to make a fool out of myself. Like I know I am going to get hung up on errors with the estimation of parameters with prior distributions using Markov Chain Monte Carlo, to start off with.
Anyways, I need to write a list with time spans so I can focus my approach
Scott Leibrand
@scottleibrand
Jan 18 2015 22:49
I keep hearing excuses. In this space, contribution comes from doing: writing code and sharing it. Just talking about what you want to do, or why you can't/haven't done it, just wastes everyone's time.
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:49
I don't want my work to be messy. I have to make comments.
Ben West
@bewest
Jan 18 2015 22:49
our works is messy
Scott Leibrand
@scottleibrand
Jan 18 2015 22:50
The key is to get something built they works, even in a basic/limited context.
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:51
I care about the quality of my work. I have a lot of things going on that impede on my work, and I am lucky to be semi-productive due to my health situation recently. I am not going to publish something if it doesn't work the way it is intended to work.
OK
Ben West
@bewest
Jan 18 2015 22:51
more limited, smaller in scope the better
Scott Leibrand
@scottleibrand
Jan 18 2015 22:51
It's fine if it's messy, only works in certain situations, or whatever. That's still progress.
Ben West
@bewest
Jan 18 2015 22:51
repo that only generates predictions cannot by dangerous
Scott Leibrand
@scottleibrand
Jan 18 2015 22:52
There are lots of us here who do nothing more than make improvements to others' work. That's the beauty of open source: you don't need to have a finished product for it to be useful.
Ben West
@bewest
Jan 18 2015 22:53
also, the smaller and more isolated, focused the code base, the better
Scott Leibrand
@scottleibrand
Jan 18 2015 22:53
You'll find that when you start collaborating and sharing code, the whole dynamic changes.
Ben West
@bewest
Jan 18 2015 22:53
everyone loves little libraries with only 2 or 3 functions to configure and do exactly one thing
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:54
OK, well there's not a lot of things that I can omit in the actual AP models, truthfully.
I will contribute then
Ben West
@bewest
Jan 18 2015 22:54
but coding up just a model
no ui or anything
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:54
I don't like UI
Ben West
@bewest
Jan 18 2015 22:54
just the model is inherently not dangerous at all
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:54
I don't usually do OOP.
Ben West
@bewest
Jan 18 2015 22:54
can be functional or whatever
once you publish something, be prepared for critical feedback, btw, everyone's a critic
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:55
OK
Scott Leibrand
@scottleibrand
Jan 18 2015 22:55
One thing you could do is look at the functions in wip/iob-cob that do predictions. They're crap, but they work.
You can undoubtedly improve some parts of them.
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:56
Alright, I will do so right now.
Scott Leibrand
@scottleibrand
Jan 18 2015 22:56
Figure out what we're doing poorly that
Ben West
@bewest
Jan 18 2015 22:56
I was playing with svg gradient and filter effects
Scott Leibrand
@scottleibrand
Jan 18 2015 22:56
would benefit from some application of work from the literature.
Ben West
@bewest
Jan 18 2015 22:57
it's possible to smudge those dots, casting a shadow of cob/iob
so idea is each dot would be smudged with a cone of confidence with opacity, shape, color gradients to help show range/direction of forces each dot is undergoing
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:58

The one go to resource would be this. It's the closest thing to an agreed upon model for BG prediction:
A Novel Adaptive-Weighted-Average Framework for Blood Glucose Prediction

But, I will look for improvements for starters

Ben West
@bewest
Jan 18 2015 22:58
so they are close enough generally, that the shadows would overlay
but needs some trig I think to describe how to point the shadows according to iob/cob
I have no idea how to do that
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:59
what kind of trig?
Scott Leibrand
@scottleibrand
Jan 18 2015 22:59
Cool. Just remember, the goal is not a perfect model. It's something more accurate than we have today.
Ben West
@bewest
Jan 18 2015 22:59
I'm terrible at maths, and a lot of this vis requires it
or just on parity with what we have today
diabeticgonewild
@diabeticgonewild
Jan 18 2015 22:59
Got it.
Ben West
@bewest
Jan 18 2015 22:59
anything that can accurately report failings of status quo is also good
diabeticgonewild
@diabeticgonewild
Jan 18 2015 23:00
Like sensor failure?
Ben West
@bewest
Jan 18 2015 23:00
there's a general belief that what we have on the market is basically safe and gets the job done well
and that's why it's on the market
so by status-quo I mean measuring what actually happens, like even basic things like number of hypos/hyper events expected if someone undergos recommended treatment
Dana Lewis
@danamlewis
Jan 18 2015 23:03
@bewest, @scottleibrand and I got retro auto reports of what loop is doing. Still some tweaks needed, but awesome already :) going to help a lot wIth overnight basal tweaking
Ben West
@bewest
Jan 18 2015 23:04
yes
yeah, I think a lot of the individual "features" actually need to be their own projects/repos
anything that would get tested on it's own better, but long road on that ;-)
in short term anything that works, we can just patch patch patch
Dana Lewis
@danamlewis
Jan 18 2015 23:05
Haha instead of our closed loop blob of awesomeness? :) we like to add to our blob, then add piece by piece.
Ben West
@bewest
Jan 18 2015 23:06
yes, that's good for investigation, quick prototypes
Dana Lewis
@danamlewis
Jan 18 2015 23:06
Yup
Ben West
@bewest
Jan 18 2015 23:06
and to find the stuff worth keeping
once we identify the stuff worth testing/keeping
Dana Lewis
@danamlewis
Jan 18 2015 23:06
Anyway, will post screenshots in a day or two to show ya'll since gitter phone doesn't enable.
diabeticgonewild
@diabeticgonewild
Jan 18 2015 23:15
So this is a second order linear regression?
Ben West
@bewest
Jan 18 2015 23:15
think so
diabeticgonewild
@diabeticgonewild
Jan 18 2015 23:15
at least for IOB?
interesting.
Scott Leibrand
@scottleibrand
Jan 18 2015 23:19
Monica: is what 2o LR?
diabeticgonewild
@diabeticgonewild
Jan 18 2015 23:20
Line 1515: var iobContrib=treatment.insulin(1-0.001852xx+0.001852x);
Scott Leibrand
@scottleibrand
Jan 18 2015 23:23
Ya, IOB formula is two quadratic formulas
One for before peak, another for after.
diabeticgonewild
@diabeticgonewild
Jan 18 2015 23:23
Nice, not a bad idea for simplicity at all.
Scott Leibrand
@scottleibrand
Jan 18 2015 23:23
@kenstack is working on getting a better set of formulas based on Walsh.
diabeticgonewild
@diabeticgonewild
Jan 18 2015 23:24
What should I start out on first, COB or IOB
I don't want to stomp on anyone's feet.
Or just prediction for BG
Chris Oattes
@cjo20
Jan 18 2015 23:24
work on whatever you like I guess
diabeticgonewild
@diabeticgonewild
Jan 18 2015 23:24
Prediction for BG might be the way to go, given @kenstack
Scott Leibrand
@scottleibrand
Jan 18 2015 23:27
Would be nice to improve the way we combine AR and IOB/COB predictions.
Right now I start with 100% AR then switch to iob/cob gradually over 30m.
Might be a place for you to implement your weighted averages idea.
Ben West
@bewest
Jan 18 2015 23:29
or coming up with the js to cast iob/cob shadows off the dots
diabeticgonewild
@diabeticgonewild
Jan 18 2015 23:30

OK, will do AR.

@bewest , I don't know a lot of OOP. Data structures is more of my thing...

It would be better use of everyone's time
Scott Leibrand
@scottleibrand
Jan 18 2015 23:30
Sounds like Monica would prefer to focus on models and predictions vs UI
Chris Oattes
@cjo20
Jan 18 2015 23:31
@scottleibrand what is "AR"?
Scott Leibrand
@scottleibrand
Jan 18 2015 23:31
And, Seahawks are in the super bowl! :-)
diabeticgonewild
@diabeticgonewild
Jan 18 2015 23:31

It's not a preference. I am just good at math.

Yay! I always root for the Seahawks...I'm from Seattle.

Scott Leibrand
@scottleibrand
Jan 18 2015 23:31
Autoregressive.
nightscout has a second-order AR model for predicting BG based on current slope.
It's what is used to make the blue cone, and for the predictive alarms.