loudnate / openaps-predict Goto Github PK

In support of #17 etc., we should add the DIYPS IOB curve formula. For 3h DIA, it uses a bi-linear activity curve that goes from 0 to peak at 75m, then back down to 0 at 3h. The resulting IOB curve is curvilinear, and looks a lot like the Walsh and Medtronic curves. Those curves can then also be scaled to all other DIAs just by scaling the X axis.

https://github.com/openaps/oref0/blob/master/lib/iob/calculate.js
https://github.com/nightscout/cgm-remote-monitor/blob/master/lib/plugins/iob.js

To support additional DIA values (as well as partial-hour lengths). This will also address #17.

Add initial glucose momentum estimation and enable the easy addition of other short-term predictive models. First pass can simply be a linear extrapolation, for example.

Add predictions (and subsequent downstream calculations) for DIA = 2 h for those with the shortest available DIAs in the minimed pumps.

Would be nice to have an --at option to get just an instantaneous IOB reading as of a provided timestamp (likely usually from read_clock).

Much like meal markers are used in combination with the bolus wizard entries for COB analysis, it would be great to use the exercise marker to adjust the prediction. There are many ways to do this, but one is to predict a drop of x mg/dL over y min for each marker that is placed. Perhaps set it at a 1/2 U ISF equivalent or something (all could be user-configurable). When I'm about to bike or something, I can simply add a marker or two on the pump, which will shift the prediction and adjust accordingly. If I'm exercising longer, I can add more markers.

Might only apply to x23 series, not sure.

This is to support NS API responses

Just because we have a glucose report doesn't mean its contents are "recent", e.g. the sensor could have gone bad or out-of-range.