RunFitter

A tool for fitting the output of a simulation to data. Simulation output variable(s), and parameters can be specified by the user. This tool may be saved in a session. This tool uses the fit_praxis() method and the least squares error function calls run().

The minimization function used by the fitter calls the hoc “run” procedure (see InitRun). The output simulation variable data is stored using the Vector.record() function, ie values are copied from the variable to the vector at the end of finitialize() and at the end of fadvance() whenever t passes the x values of the data. Fitting parameters are varied using one element vectors registered with Vector.play(), i.e., the value is copied from the vector to the parameter at the beginning of finitialize(). The notion of a fitting parameter has been extended so that one can call an arbitrary statement so that the fit value can be used to assign values to a collection of hoc variables. E.g., globally setting a range variable.

Any number of data weight regions can be specified in order to ignore artifacts weight critical regions more heavily.

Fitting parameters and output parameters are registered with the play/record lists only when the “Running” checkbox shows that the widget is executing.

Warning

Multiple instances of the RunFitter widget can be present. But make sure you are not Running more than one at a time. When saving a session involving it is necessary that on retrieval the necessary variables exist that are used by the fitter. In the case of extra fit variables this means that the master fitter should be selected prior to the slave fitters when using the Print & File Window Manager.

When a parameter is very close to 0, its limited resolution in a field editor may cause problems. In this case define the parameter to be a scaled version of the actual desired value, e.g.,

g_pas = 0.0001*$1

Note

This scaling rule is a HOC expression, not a Python or MATLAB expression. You must use a HOC expression in this GUI tool, even if you are using a different programming language.

Only change morphology parameters such as diam and L using a statement involving $1. Otherwise the system will not be notified that diameter is changing.

ReadData

Pops up a filechooser for reading the data file. The first number in the file is the number of points. Subsequent pairs of numbers are x and y values of the data. Alternativly the Graph menu can be used to invoke the DataFromClipboard item (see RetrievefromFile).

When data is read from the clipboard, that data is saved when the tool is saved in a session. However, if the ReadData button is used the filename is saved.

CurrentValuesAsDefault

Any checkmarks are removed from the default field editors in the parameter list.

Variabletofit

Pops up symbol chooser. The syntax of the variable must be in a form which is a valid argument to a Vector.record(var) function. Practically speaking, this means that if the variable happens to be a density range variable then the entry string must contain an explicit section arc length parameter. e.g., soma.v(.5) (Note: No matter what programming language you are using, you can only use HOC expressions with the RunFitter tool). Point process variables can use either an objref prefix or the internal object name, e.g., SEClamp[0].i . Navigating to a variable name with the chooser generally yields a valid name. If more than one variable is to be fit to separate data curves, invoke a slave RunFitter with from the Extras/AnotherVariableToFit menu item.

Parameterstovary

Every time this button is pressed it pops up a symbol chooser for appending a variable to the list of parameters to be varied in order to least squares fit the “Variabletofit” to the data. DefaultFieldEditor’s for these parameters appear in the top right box of the widget. These parameters must be in a form acceptable to the Vector.play(var) function. ie density range variables must contain an explicit arc length parameter.

In the case of a simulation consisting of more than one compartment, it is often necessary to identify a parameter with a set of values. In that case one can enter an arbitrary HOC statement (in general, anything entered into GUI fields must be HOC expressions not Python or MATLAB) involving the parameter “$1”, e.g.,

forall g_pas = $1

Note

This field requires HOC not Python or MATLAB expressions.

Each parameter has a checkbox to the left of its name. When checked, the value will be adjusted during a fit to optimize the model to the data. If not checked the parameter will be held constant during the fit.

Extras

DataWeights

Pops up a panel showing the boundaries and weight values for each data region. The boundaries can also be manipulated by selecting the AdjustWeightRegions tool from the Graph menu (right button) and then dragging the boundary lines. Weights are defined so that data points a small region will have a total weight equal to the data points in a large region when the interval weight values are the same.

ParameterRanges

Allows specification of the allowable parameter range for a fitting parameter. If praxis uses a parameter outside this range, the least squares error function will return 1e6 without calling the run procedure. Default parameter ranges for all fitting parameters are initialized to 1e-6 to 1e6

Changeparmfromlist

Pops up a browser with all parameter names. Double clicking on a name will pop up a string dialog which can be used to change the parameter name or statement.

Removeparmfromlist

Pops up a browser with all parameter names. Double clicking on a name will remove that parameter from the list.

SaveRestoreFitParams

The SaveFitParms menu item saves the current values of parameters, parameter range limits, and and whether the parameter is to be held constant during a fit.

The SaveFitBrowser menu item pops up a list browser. Double clicking on these items will copy the saved parameters etc, back into the current parameter panel.

NumberOfDataRegions

Select the number of data regions to use in weighting the data.

AnotherVariableToFit

Pop up a slave RunFitter to allow simultaneous fitting of several sets of data to several fit variables. A RunFitter Slave does not have a parameter panel but has independent selection of data, variable to fit, and data weight regions.

When saving a RunFitter Slave to a session, it must be placed on the paper icon of the PrintWindowManager AFTER its master.

Dofit

Calls praxis to do the fit. During a fit, intermediate results are occasionally printed to the xterm window showing the progress of the fit. While the widget is working the Running checkbox is checked. If the StopatnextQuadForm button is pressed while the fit is running, the fit will stop at at the end of its current main loop returning its current best fit along with a print of the principal axes and principal values. Left alone, praxis will return when it is within 1e-5 of the local minimum. If Dofit is pressed while the StopatnextQuadForm is checked praxis will stop after one main loop (calculate principal axes and values)

SingleRunFit

Call the least squares error function once. This results in a single simulation run with the parameter values displayed in the panel. The “Error of fit” field editor shows the square norm of the data - outputvariable.

StopatnextQuadForm

Cause praxis to stop after it finishes its current principal axis/value computation. Be patient, it may be necessary to wait for several runs before the computation completes. To immediately quit, press the Stop button on the RunControl. This will stop the fit immediately and set parameters to the best fit found so far. Only do a ^C if necessity demands and then remove the check by pressing SingleRunFit.

Running

Checked when in the process of doing a Dofit or SingleRunFit. When checked one should not try to change the widget by changing parameters or doing a recursive run. The check may not be accurate if the previous run generated a runtime error since the check is removed only if the call to praxis returns normally. In this case one may press the SingleRunFit button and follow the instructions to remove the check.

Roughfit

Uses just 4 points per weight region to fit the data. This can allow the variable time step method to run much faster since there are many fewer recording events requested.

BeQuiet

Turns off printing by the praxis function and does not flush the graphs after run() is called.