These are chat archives for nightscout/ios

27th
Sep 2015
kenstack
@kenstack
Sep 27 2015 00:03
Delta is right below bg - in the iOS companion app you can choose whether dex or raw is primary
Pete
@someoneAnyone
Sep 27 2015 03:20
@scottleibrand are you back up and running?
Scott Leibrand
@scottleibrand
Sep 27 2015 03:22
Yep thx.
Jason Calabrese
@jasoncalabrese
Sep 27 2015 03:45
looks great @kenstack
Pete
@someoneAnyone
Sep 27 2015 13:27
@kenstack , help me get some of that into the phone side! :-)
kenstack
@kenstack
Sep 27 2015 14:36
@someoneAnyone yes I absolutely will - sorry to be offline so much - will be back to this mostly full time in a week or 2 - my supposedly part time day job turned full time for a month ...
Pete
@someoneAnyone
Sep 27 2015 14:37
Its cool! @kenstack I’ve been busy too… my wife was traveling for work so I’ve been runing ragged lately.
@kenstack , I’m playing with iOS-Charts at the moement. Have you played with it at all?
Simulator Screen Shot Sep 27, 2015, 10.40.59 AM.png
I still have a ways to go
Pete
@someoneAnyone
Sep 27 2015 14:42
I’m also working on a re-write that uses more cocoapod libraries making my code smaller.
kenstack
@kenstack
Sep 27 2015 21:32
in https://github.com/Perceptus/nsar.git I added a file called nsvsmooth that now compares the smoothed linear velocity prediction to actual dex data - you can change the prediction time ahead with a slider - just add the name of your site in the top input box. At 20 minutes its really good - at 30 its still quite accurate on a drop or rise but if you have oscillating BGs on the way up or down at 30 min you definetely get some false positives or negatives on the way. My goal with all this is for my ap re predicting lows asap - I think its quite good at predicting when youll go low in the next 30 min even if some of the predictions are a bit off in BG - Ive been using it in my alarm SW and Im really pleased so far but Im very interested in feedback. The smoothing I use is a simple regression type smoother which Ive tested against Kalman, Fourier, and other simple low pass filters. Very little lag and it greatly smooths the velocity. Ive tested it against both raw and against actual dex too. It really helps a lot on raw given the jumps
Scott Leibrand
@scottleibrand
Sep 27 2015 21:56
Nice. I use a simple 15m moving average extrapolated out 15m in openaps, but would love to see some better algorithms I could plug in instead.
I use the 15m projection to calculate the deviation from BGI, and add/subtract that to/from eventualBG.
if we could reliably extend that out to 20-30m that would help a lot with being proactive enough for temp basals to ward off lows and help more with meals.