Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
  • Jul 04 2017 03:25
    @scottleibrand banned @Prosulpump
Eric
@ecc1
but there are historical reservoir levels in NS, so one could write an auxiliary checker that flagged inconsistency between event-based delivery calculation and reservoir delta (not volunteering)
Scott Leibrand
@scottleibrand
yes, you could do that (also not volunteering)
Eric
@ecc1
where are the interns when you need them
Scott Leibrand
@scottleibrand
haha yeah
tzachi-dar
@tzachi-dar
Thanks, that explains it.
David Klapan
@dklapan_twitter
It seems that error regarding "Couldn't invoke_pumphistory_etc" is happening again. I've seen on histoy of this gitter that it happened already. How did you managed to solve it?
Listening for 3s: .No interfering pump comms detected from other rigs (this is a good thing!) Continuing oref0-pump-loop at Wed 14 Aug 07:44:11 CEST 2019 Preflight (x12 models do not support SMB safety checks, SMB will not be available.) OK. Profile less than 60m old; Profile valid. Pump history update failed. Last record 2019-08-13T12:43:57+02:00 Couldn't invoke_pumphistory_etc - continuing Retry 1 of refresh_pumphistory_and_meal Pump history update failed. Last record 2019-08-13T12:43:57+02:00 Couldn't invoke_pumphistory_etc - continuing Listening for 1s: .No interfering pump comms detected from other rigs (this is a good thing!) Continuing oref0-pump-loop at Wed 14 Aug 07:44:48 CEST 2019 Retry 2 of refresh_pumphistory_and_meal Pump history update failed. Last record 2019-08-13T12:43:57+02:00 Couldn't invoke_pumphistory_etc - continuing Listening for 3s: .No interfering pump comms detected from other rigs (this is a good thing!) Continuing oref0-pump-loop at Wed 14 Aug 07:45:06 CEST 2019 Retry 3 of refresh_pumphistory_and_meal Pump history update failed. Last record 2019-08-13T12:43:57+02:00 Couldn't invoke_pumphistory_etc - continuing Couldn't refresh_pumphistory_and_meal oref0-pump-loop failed. pump_loop_completed more than 15m old; waiting for 9 s silence before mmtuning "HAT Display Updated" Listening for 9s: .No interfering pump comms detected from other rigs (this is a good thing!) Continuing oref0-pump-loop at Wed 14 Aug 07:45:18 CEST 2019 Listening for 9 s silence before mmtuning: Listening for 9s: .No interfering pump comms detected from other rigs (this is a good thing!) Continuing oref0-pump-loop at Wed 14 Aug 07:45:27 CEST 2019 mmtune: "868.400", 3, -71 -- "868.450", 3, -84 waiting for 48 second silence before continuing Listening for 48s
Scott Leibrand
@scottleibrand
How close is the pump to the rig? Is preflight or anything else failing? Do you have mmtune results?
If nothing else is failing, sometimes you can jump start it by deleting pumphistory and forcing it to pull a full fresh one. It’ll eventually do that on its own though.
(So it has probably done so by now)
13916953461
@13916953461
@scottleibrand Hi, can I ask the meaning of the predicting carb impact and the remaining carb impact. Why their shapes are different?
Jens Heuschkel
@juehv
just discovered that a web interface sneaked into the standard installation. Is there any chance to turn this off by a preference option ?
Scott Leibrand
@scottleibrand
Why do you want to turn it off?
@13916953461 predCI comes from actual observed carb absorption, decayed to zero over time based on COB. If current observed CI is too low for predCI to predict absorption of all the COB over a reasonable timeframe, we predict that the remaining COB will decay, with CI ramping up from current levels to a peak before declining to zero.
Jens Heuschkel
@juehv
@scottleibrand the pi rig is running at 400% CPU load and I want to turn off unneeded services.
Would be nice to come down to 100% load :sweat_smile:
David Klapan
@dklapan_twitter
Yep it worked by itself, but now it's stuck again. Pump is about 3 meters at furthest, in same room. How can I jumpstart it usually?
Eric
@ecc1
@juehv I haven't looked closely, but if the server isn't used and is just listening on a port, it shouldn't be using any cycles
Jens Heuschkel
@juehv
@ecc1 year maybe true and ram is not an issue at the moment... I already found out that the pump-loop script produces 380% of the 400% load
Jens Heuschkel
@juehv
I built a new text based display screen. Any feedback is appreciated! You can find it here: https://github.com/juehv/openaps-menu
It can be activated by using the preference "status_screen": "textstatus"
It is designed to update independendly every minute. So I put ~/src/openaps-menu/scripts/status.sh into my cron tab.
Jon Cluck
@cluckj
nice
I've been running the graph status.sh in my cron every minute for a while
phraktyl
@phraktyl
I know this isn't quite the place for this, but has anyone had the issue of the (for lack of a better term) plug on their Medtronic pump opposite of the reservoir coming out? I've been trying to prime me pump this morning and it just pushes the back out instead of pushing the insulin up.
Scott Leibrand
@scottleibrand
Stop using the pump immediately.
That failure is a big safety risk: if it pops out, you don’t get any insulin, and if it gets pushed back in, you get all that insulin at once. Both are dangerous.
It is possible to glue it back in place, but I don’t remember the details. If you do choose to glue it, you’ll want to run it disconnected from the body for a few days pumping just water to make sure it is solid.
Or better yet, stop using that pump, use another one, and keep the glue-repaired one as an emergency backup.
phraktyl
@phraktyl
I was aid Thai was going to be the answer. Time to figure out where I put my backup!
phraktyl
@phraktyl
That Okay. Found my backup pump. It's not powering on with a known good battery. Looks like it's going to be one of those* Mondays.
phraktyl
@phraktyl
Battery contacts cleaned. One A21 error later, and it looks like this will work. Just edit the r
Rerun oref0 setup with the new serial?
Martin Haeberli
@mhaeberli
recommend indeed to run oref0-setup
phraktyl
@phraktyl
Haha! Just kidding. I thought I had a 715 as a backup pump. I have a 751. Looks like I won't be looping for a while.
Martin Haeberli
@mhaeberli
sorry to hear it!
where are you, geographically, by the way?
phraktyl
@phraktyl
Southern California
Martin Haeberli
@mhaeberli
Bay Area!
so not so far
phraktyl
@phraktyl
Nice!
13916953461
@13916953461
@scottleibrand hi. I want to ask some questions aboutthe algorithm.
1) I am not clear the reason why you use both predicting carbs impact and remaining carbs impact in the algorithm?
2) What is the logic of the algorithm about the carbs impact part?
3) For the predicting carbs impact and remaining carbs impact, they are linear. I have analysed the carbs impact part in Hovorka Model and Two-Compartmental Model, and their changes are curves, so I think they are closer to reality. Do you think it is possible to replace the linear line change by curves through replacing some lines of codes?
I will appreciate if you can solve my confusion
Scott Leibrand
@scottleibrand
when you first enter carbs, and there are no positive deviations, there is no current carb impact, so the prediction based solely on CI would never decay all the COB. So we instead assume that deviations will increase from the current low level to a peak, and then back down to zero. The height of the peak is calculated to absorb all the carbs on board.
If you wanted to replace my triangles with curves you could do so, but I’m not sure how you would ensure that such curves are more accurate than a simple linear approximation.
But if you do manage to come up with a method for generating better postprandial BG predictions and/or for dosing for carbs, we have some good techniques for evaluating whether such an algorithm change would be safer and more effective, using backtesting against real world data.
Jens Heuschkel
@juehv
@scottleibrand That sounds interesting! Can you tell me more about the evaluation techniques (since we are working on BG predictions in our lab)?
Brian Quinion
@twain47
@scottleibrand I'd be interested in that too. I'm trying to make some tweaks and struggling to find ways to effectively evaluate the result of changes.
Scott Leibrand
@scottleibrand
The simple-simulator branch of oref0 has the ability (with some manual glue) to run oref0 in backtesting mode. The way it works is: first you run autotune on someone’s NS, which captures in its logs the observed deviations for each 5m interval of the day, which represents what BG actually did, relative to what would gave been expected based on oref0’s insulin model. Then, you can apply that sequence of deviations to a simulated run, such that the simple-simulator calculates what BG would have been, doses insulin in response to that, and then repeats the calculation for every 5m interval. That gives you a scenario of what would have happened differently if you had doses insulin differently, by backing out the expected impact of the insulin that was actually dosed, and replacing it with the expected impact of different dosing. It implicitly assumes that deviations are independent of insulin dosing, which requires that the insulin model be already accurate, and that the BG impact of carb absorption would be unchanged if you tweaked dosing. That’s a fairly strong assumption, but in situations where it holds, it allows you to evaluate incremental changes to the dosing algorithm against data collected from someone who got a slightly different amount of insulin, and thereby determine if such a tweak likely would have improved or worsened their postprandial outcomes.
13916953461
@13916953461

@scottleibrand Thank you for your explanation. I still have some confusion. When someone finish a meal, the carb impact cannot be detected immediately, so we assume the carb impact will change like triangles. But when will the predicted carb impact (linear decay) appear?

Whether the carb impact will change like triangles at firstl, and when the carb impact can be detected, then the carb impact will decay linearly?

Scott Leibrand
@scottleibrand
If the time horizon is 4h, then we look at current CI, decaying to zero slowly enough to absorb all COB, and if if decays within 4h, great. If not, the remainder that doesn’t absorb by 4h is considering remainingCarbs and its absorption is calculated as increasing to a peak (say at 90m) and then decaying to zero by 4h.
There are figures in the docs that help explain this.