nightscout / AndroidAPS

Opensource automated insulin delivery system (closed loop)
https://wiki.aaps.app
GNU Affero General Public License v3.0
725 stars 1.76k forks source link

Dev-b AutoISF / Can't calculate acceleration #3413

Open robertrub opened 2 months ago

robertrub commented 2 months ago

I switched G6 transmitters which resulted in a desynchronisation of the 5 minutes interval which started the recalculated values.

Screenshot_20240824_212910_AAPS.jpg

I think (just a guess) that AutoISF plugin does not like recalculated values and can't get a "reliable" acceleration; all acceleration numbers are at 0.

As it didn't see the deceleration, it selected an agressive ISF for those 3 last BGs (were over target).

Screenshot_20240824_221303_AAPS.jpg

As you can see, just before the drop, there was a dark dot.

Here are the logs (around 22h13 and before) AndroidAPS_LOG_1724530521674.log.zip

ga-zelle commented 2 months ago

There are only 4 loop events in a single log file. It would help if under Maintenance you disable WEAR (because it just jams the file and makes reading very hard) but enable GLUCOSE which I think is the important one here.

robertrub commented 2 months ago

Ok, did the changes to wear and glucose last night. Here are the new logs. There are still dark green dots but I don't know if acceleration was at 0 or not.

Screenshot_20240825_080246_AAPS.jpg

AndroidAPS_LOG_1724565734883.log.zip

robertrub commented 2 months ago

Calculated BGs (transmitter change 24h ago)

Screenshot_20240825_082453_AAPS.jpg

Acceleration at 0

Screenshot_20240825_082426_AAPS.jpg

robertrub commented 2 months ago

OK, I found 2 cases in the logs where Parabola settings change to 0 after a BG time recalculation and the BGs go into a "loop" with 3 minAGO coming over and over. In the first ZIP, it is 30, 40 and 45 min ago. Look before 9:16:34:840 AndroidAPS._2024-08-2500-00-28.93[1].zip

robertrub commented 2 months ago

Recalculated data period is over. There are no more dark green dots on the graph and AutoISF is working normally once again.

There is a problem with the recalculated data and AutoISF.

ga-zelle commented 2 months ago

The problem is with gaps in data, not recalculated ones. AutoISF deals with them as expected. Without G6 I have recalculated all year round and no such problems..

robertrub commented 2 months ago

Sorry but I disagree. As soon as the yellow triangle disappeared (and the dark green dots too), the problem of parabola at 0 stopped. Field test says that I'm right ;)

The recalculated data of Libre and G6 seems to be different. No idea what difference. Maybe the difference is that Libre has just seconds of difference when a transmitter change makes minutes of difference...

Another possibility : Are you using the AutoISF plugin in AAPS dev or are you using your "own" AutoISF build ?