This project has moved and is read-only. For the latest updates, please go here.

DataSet

May 1, 2011 at 1:36 PM

Hello

I'm using the first time this library. Could it be that the documentation does not longer fit to the library version 2.0 ?
I tried to fit some data and used your example in the Statistics documentation, but the library does not any longer have the class "DataSet" ? Am I wrong?
Is this not longer required? I worked with the UncertainMeasurement class to add xy value pairs. For fitting it is necessary to set the dy value not to zero. A zero value
returns an exception in the FitResult. Is this true? Are there any other DataSets to use for fitting where dy is not relevant?

Are there any other tutorials for this library?

Greetings

May 1, 2011 at 10:37 PM

Hi Mariusmeir,

You are right that in version 2.0 DataSet has been renamed to UncertainMeasurementSample; we need to update our documentation page! But even before version 2.0, this class was intended to represent a sample of data points with uncertainties (dy). The classes for data without uncertainties are BivariateSample (for x/y pairs) and MultivariateSample (for x[]/y pairs). They have some regression functionality (BivariateSample.LinearRegression, BivariateSample.LinearLogisticRegression, MultivariateSample.LinearRegression), but they do not yet support regression to an arbitrary, user-defined function like UncertainMeasurementSample does. If that's what you need, you have two choices:

1. You can wait for version 2.1, where we do intend to add methods for regression to an arbitrary, user-defined function to BivariateSample and MultivariateSample.

2. You can actually do this right now using by purposely "mis-using" the UncertainMeasurementSample. Just set all uncertainties to the same value, say 1.0. Then all points will be equally weighted and the calcualted parameter values will be the same as you would from a least-squares regression without uncertainties. To get the right parameter uncertainties, you just need to re-scale the calculated uncertainties by the square root of the chi-squared value (FitResult.GoodnessOfFit.Statistic).

Thanks for your inquiry. Feel free to let us know anything else you find confusing/wrong/missing from the library, either in your reply or by creating work items in the issue tracker pane.