Figure of Merit for Calibration

[lorduintah]

John -

In statistics, usually when some data is fit to a model, there is an associated figure of merit. For example a least squares fit of data to a straight line, one can find the coefficients that best describe the line fitting that data, but also a figure of merit - usually R^2. So the method derived to extract coefficients usually has embedded a way to judge the quality of the parameter estimations.

Since calibrating ride data to yield the parameters that are subsequently used to generate power values - instantaneous and overall = does something like ISAAC have that capability - if so, it would be useful to really allow users to evaluate how good their ride data was during calibration. This would provide feedback as to the quality of their calibration. Furthermore, using data from many rides would allow one to collate the multiple regression solution to the parameters.

A lot of data points are collected during a ride - calibration or otherwise. So the comparison of the expected data to the actual could provide something like what I have outlined.

Thanks,

Tom

[Velocomp]

Sorry for the delay in responding.

We used to do R^2 calculations in Isaac, to determine the quality of fit between PP and DFPM data. This was in a secret menu...

We got really good results: R^2 values routinely above 0.8 (this is by my memory...)

Alas, this information is not currently accessible in the released version of Isaac, and the person who developed Isaac is no longer with us. So, I'm afraid this will remain a "some day" item.

[lorduintah]

Fair enough.

I would also guess that the overall model - fitting over a range of wind, slope, acceleration, etc. is not a simple linear or quadratic case - this is a multidimensional situation - but I wonder if the wind calibration, alone is a tad more straightforward. Just a math geek's curiosity.

However comparing Isaac estimates for power to a DFPM would make that R^2 meaningful - at least in a comparison of the two to each other.

Tom