overriden calibrations used in slopeOOBHandler ?

[JohanDegraeve]

when overriding a calibration, the latest calibration is changed so that sensor_confidence=0 and slope_confidence=0 (see https://github.com/StephenBlackWasAlreadyTaken/xDrip-Experimental/blob/master/app/src/main/java/com/eveningoutpost/dexdrip/Models/Calibration.java#L664)

Then when a new calibration is created, a call is made to calculate_w_l_s
this one makes a call to slopeOOBHandler
slopeOOBHandler uses the latest 3 calibrations. For this it calls Calibrations.latest(3)
https://github.com/StephenBlackWasAlreadyTaken/xDrip-Experimental/blob/master/app/src/main/java/com/eveningoutpost/dexdrip/Models/Calibration.java#L550

But latest(3) is picking up all calibrations with matching sensor id, the latest 3, that means it also takes the latest one which is actually being overriden.

Is that normal ?
I would think that it's better to get Calibration.allForSensorInLastFourDays and then take the latet 3 from that result.

[AdrianLxM]

I's day: forget about that if you want to have the same behaviour as xDrip:

As you have already pointed out here: #361 , there is another bug that estimate_bg_at_time_of_calibration is never written.
Therefore the only cases where one of the latest calibrations would be used
(e.g. https://github.com/StephenBlackWasAlreadyTaken/xDrip-Experimental/blob/master/app/src/main/java/com/eveningoutpost/dexdrip/Models/Calibration.java#L554) can never happen.

Basically what stays it is just a distinction if the initial calibration is done (then we have at least 2 calibrations) or if a third one was entered and then return a default value based on the default parameters and sensor age.

The whole calibration algorithm would need an overhaul.

[JohanDegraeve]

returnvalue would be different depending calibrations.size()

if L544 will never happen as you say, then the value would be

if size = 3
Math.max(((-0.048) * (thisCalibration.sensor_age_at_time_of_estimation / (60000 * 60 * 24))) + 1.1, sParams.getDefaultLowSlopeLow());

if size = 2
Math.max(((-0.048) * (thisCalibration.sensor_age_at_time_of_estimation / (60000 * 60 * 24))) + 1.1, sParams.getDefaultLowSlopeHigh());

The test I'm doing now is doing an override immediately after initial calibration.
I'm indeed trying to get exactly the same behaviour as the Android version. Actually I had an error that I was taking the latest3 with sensor confidence != 0 and slope confidence != 0, so I was getting size 2 while the Android version is getting size 3.

But ok I'll align it with the Android version

[AdrianLxM]

Yea, sure. Overriding the initial calibration is an exception.

After the initial calibration, the slope usually is 1.15 and after
another one (or overwritten one) 1.08.
Freshly started sensors don't have such high slopes anyways for me and
I've corrected those values down in my version.

But actually it is both a situation where we don't have much information
so setting a default slope close to 1 and adjusting the intercept seems
reasonable to me.

The inicial calibration and how it is represented and overwritten also
is a bit... well. Actually just one bg entry gets overwritten.

Overall this is just the slopeOOBHandler. Usually linear regression does
determine slope and intercept. Just if the slope is too off, reasonable
defaults are returned. They are just general estimations and will be off
for an individual system/patient anyways. I use individual ones that are
much lower - and widened the range where the linear regression result is
not overruled and go very well with it.

On 20/08/16 23:26, Johan Degraeve wrote:

returnvalue would be different depending calibrations.size()

if L544 will never happen as you say, then the value would be

if size = 3
Math.max(((-0.048) * (thisCalibration.sensor_age_at_time_of_estimation
/ (60000 * 60 * 24))) + 1.1, sParams.getDefaultLowSlopeLow());

if size = 2
Math.max(((-0.048) * (thisCalibration.sensor_age_at_time_of_estimation
/ (60000 * 60 * 24))) + 1.1, sParams.getDefaultLowSlopeHigh());

The test I'm doing now is doing an override immediately after initial
calibration.
I'm indeed trying to get exactly the same behaviour as the Android
version. Actually I had an error that I was taking the latest3 with
sensor confidence != 0 and slope confidence != 0, so I was getting
size 2 while the Android version is getting size 3.

But ok I'll align it with the Android version

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
#378 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/AJPmwioP336nAbUnQ5NW0CiV0Pwqcnm7ks5qh3EfgaJpZM4JpKQp.