CurveExpert Basic Documentation

Table Of Contents

Previous topic

Contacting the Author

Next topic

References

This Page

Frequently Asked Questions

Can CurveExpert work with more than one independent variable?

Unfortunately, no; at this time, CurveExpert only supports one independent variable. For multivariate support, check out CurveExpert Professional.

How can I cite CurveExpert in a publication?

Hyams, D. G., CurveExpert software, http://www.curveexpert.net, 2010.

Why does CurveExpert suddenly erase my curve fits off of the ranking chart?

Any time that CurveExpert detects that the data set has changed, all curve fits are marked as invalid. The ranking chart shows only those fits that are valid, so the ranking chart is erased. The data set can change by manual editing or using the built-in data manipulation tools in the Data menu.

How can I increase the speed at which the graphs are updated?

Particularly if one has a large number of data points, the graphing windows are sometimes updated very slowly, as the data point marker must be drawn for every point. However, CurveExpert has a setting in its application defaults, called the Render Level, that can be used to ease the pain of slow redraws. The tradeoff is that a large chunk of precious GDI memory will be used for each graph is open. Choose File|Preferences, and select “Fast Rendering.” Note that when drawing the plot for the first time, or when resizing the window, it will take just as long for the graph to be drawn. The advantage comes about because once this drawing is done, it is saved. Now every time the window is covered/uncovered by another window, or moved to a different location, it does not have to be redrawn from scratch.

Why does the chi-square reported in the Information dialog not match the standard error?

The standard error and the chi-square values are very similar, but defined differently. The chi square is defined as X2 = sum((y-yi)^2/sigmai^2), while the standard error is defined as S = sqrt(sum((y-yi)^2/sigmai^2)/(N-NP)). So, if you do the math, you get S = sqrt(X2/(N-NP)).

Why is there a huge number for standard error and correlation coefficient sometimes? Is this a bug?

No, this is not a bug. This is CurveExpert’s way of telling you that your data set is badly out of scale – in the course of computing the standard error of the estimate and correlation coefficient, an accumulation of differences between the curve fit and data points and the mean and the data points takes place. If this calculation overflows, which is quite likely when the magnitudes of the x or y data sets are very large, then CurveExpert does the best it can by setting this sum to the maximum allowed by the data representation (single precision or double precision). This usually leads to unrealistic values for the standard error and correlation coefficient. The solution to this problem is to rescale your data to a magnitude of one.

As nonlinear regression progresses, the indicator for standard error (S) sometimes freezes, although the iteration counter keeps on ticking. What is going on?

Without getting into the internals of the Marquardt-Levenberg method for nonlinear regression (see the Theory section), suffice it to say that everything is still functioning correctly. If the algorithm detects that the standard error (chi square) increases during an iteration, then the new parameters are discarded. This naturally forces the chi square for the current iteration will be the same as the last. However, all is not lost – if this situation occurs, the algorithm makes adjustments to some internal parameters and performs another iteration. It is searching for an ‘optimum’ value of these internal parameters so that it can reduce the chi square and therefore find better parameters. If the chi square only adjusts itself every other iteration, then the internal parameters are ‘on the edge’ of optimum and the calculated regression parameters are being discarded every other time.

A spline fit never matches my data at the points like it is supposed to, and sometimes has spikes and/or wildly oscillating tendencies. Is there something amiss?

Two items will definitely cause this behavior – [1] fitting splines to unsorted data or [2] fitting splines to data with replicate values, i.e., two or more x data points are the same. The solution to problem #1 is easy; simply sort the data using CurveExpert’s built-in sort feature. Reapply the spline fit, and everything should work as planned. If problem #2 is the issue, then I have to question the use of a spline to fit your data. I would recommend that one or more of the replicate values be removed, or you will be forced to use a regression-type model. Other reasons that splines might fail is because of the sheer number of data points. Round-off errors in the computations can accumulate, causing incorrect splines to be shown. Be careful when applying splines to very large data sets. Again, splines are usually used on small to moderate size data sets. And as always, as yet another reminder, make sure that your data set is scaled to reasonable numbers (preferably on the order of one). High or low values may be causing excessive round-off error, overflow, or underflow in the computations.

Lagrangian interpolation sometimes bombs with a floating point exception. What gives?

Check in your data set for a replicate value, i.e., two or more x data points are the same. If this is an issue, then I have to question the use of interpolation to fit your data. I would recommend that one or more of the replicate values be removed, or you will be forced to use a regression-type model. And as always, as yet another reminder, make sure that your data set is scaled to reasonable numbers (preferably on the order of one). High or low values may be causing excessive round-off error, overflow, or underflow in the computations.

I am using a Postscript printer driver, and my plots will simply not print in the orientation (portrait or landscape) that I tell it to. Is there anything that I can do?

There is a bug in the Postscript printer driver (perhaps it is fixed by now) that causes it to always print in the orientation specified in the Control Panel. Unfortunately, this means that you cannot adjust the orientation in CurveExpert. However, you may open the Control Panel, select Printers, select the Postscript printer, and click the Setup button. Here, you may adjust the orientation. Now plots from CurveExpert will print in the orientation that you have specified through the Control Panel.

CurveExpert saves the previous window position, and I don’t want it to. How do I turn this behavior off?

From the File menu, choose Preferences. Deselect the checkbox that says “Save Window Position.”

How do I obtain a debugging report to send to the author?

The key is running CurveExpert with the -d parameter. From the Program Manager or File Manager, choose Run. Use the Browse command to look in the CurveExpert directory and find the executable, which is CURXPT.EXE. Double click on this file. You should then see the Run... dialog again with the full path to CURXPT.EXE in the Command Line field. Add a -d to the string already in the dialog, and press OK. CurveExpert will now record information both onscreen and in a file called CXPTxxxx.REC, where xxxx is a unique number. If multiple debug files are saved, they will be numbered consecutively. If you are having a problem with CurveExpert, the best way to solve the problem is to contact me, describe the problem, and send the debug file mentioned above. Obviously, the problem should be reproduced while the debug file is being made.

Why are grid lines drawn as solid lines (instead of dotted) on some computers?

I wish that it was a bug in CurveExpert; if it were, I could fix it. However, this happens because of a bug in the display driver installed on your computer. Contact your video card vendor or Microsoft for the latest driver. If your driver is the latest version, then you may report the bug to the author of the driver (usually the company that manufactured the video card).

The file import dialog never appears, now that I need it. How do I get it back?

Select “Tools|Options” and check the “Always show File Import Dialog” checkbox.