FastXRFLinearFit ignores "use_limit"

[woutdenolf]

In FastXRFLinearFit I see that the reference spectrum is also set with an explicit xmin and xmax value which come from the config if FastXRFLinearFit.fitMultipleSpectra did not receive xmin/xmax.

if xmin is None:
    xmin = config['fit']['xmin']
if xmax is None:
    xmax = config['fit']['xmax']
self._mcaTheory.setData(x=x, y=yref, xmin=xmin, xmax=xmax)

In ClassMcaTheory.setData I see this:

xmin = self.config['fit']['xmin']
if not self.config['fit']['use_limit']:
    if 'xmin' in kw:
        xmin=kw['xmin']
        if xmin is not None:
            self.config['fit']['xmin'] = xmin
        else:
            xmin=min(self.xdata)
    elif len(self.xdata):
        xmin=min(self.xdata)

xmax = self.config['fit']['xmax']
if not self.config['fit']['use_limit']:
    if 'xmax' in kw:
        xmax=kw['xmax']
        if xmax is not None:
            self.config['fit']['xmax'] = xmax
        else:
            xmax=max(self.xdata)
    elif len(self.xdata):
            xmax=max(self.xdata)

use_limit	Passed xmin/xmax?	Actual xmin/xmax used
True	Doesn't matter	From config['fit']['xmin/xmax']
False	Yes	Uses passed values (*)
False	No	Uses min/max of data (xdata)

(*) This is the case for FastXRFLinearFit. The passed values however are config['fit']['xmin/xmax'] when FastXRFLinearFit.fitMultipleSpectra did not receive xmin/xmax. In other words, use_limit is effectively ignored when doing fast fitting. This is a bug imo (limits are always applied).

[woutdenolf]

If the config has use_limit=False and I do not pass xmin/max to FastXRFLinearFit.fitMultipleSpectra I expect full spectrum fitting. But now it still applies the xmin/max limits from the config file. So that's a bug.

[vasole]

If the config has use_limit=False and I do not pass xmin/max to FastXRFLinearFit.fitMultipleSpectra I expect full spectrum fitting. But now it still applies the xmin/max limits from the config file. So that's a bug.

Correct interpretation although in the case of FastXRFLinearFit it was made on purpose to allow fast stack fitting from the command line. If I remember well, the user has no possibility to pass the fit limits other than via the fit configuration file.

[vasole]

So that's a bug.

It is desired behavior.

use_limits was to allow users to fix the limits irrespective of the zoomed region of the spectrum when setting up things interactively.

When doing a batch, the limits have to be used unless overwritten because there is no user interface to be used. It is internally made in McaAdvancedFitBatch and what you see is the equivalent implementation in the FastXRFLinearFit.

[vasole]

It is the explanation to your TODO why??? in McaAdvancedFitBatch

[woutdenolf]

It is desired behavior.

So if I batch fit (fast or normal) with a config file that was produced like this it takes the limits into account?

I don't understand how this could be the desired behavior. The box is unchecked so the limits are not used. That's what I would expect.

It is easier to uncheck a box than to reset the values to 0 and 4095. I have no way of setting them to None either from the GUI.

[sergey-yaroslavtsev]

It is desired behavior.

So if I batch fit (fast or normal) with a config file that was produced like this it takes the limits into account?

I don't understand how this could be the desired behavior. The box is unchecked so the limits are not used. That's what I would expect.

It is easier to uncheck a box than to reset the values to 0 and 4095. I have no way of setting them to None either from the GUI.

I had the same issue. While using this part of PyMca for batch XEOL fit. It looks that limits should not be used (box is unticked) while they actually applied. The main problem was when we tried to apply the same configuration for another batch and it fails because limits were used...

[vasole]

Again,:

• when using a batch, you have no means to uncheck a box.

• when you are in interactive mode, you have the choice to fix the limits or to to use the limits set by the zoomed region of the spectrum sent to the fitting window. That is the only usage of that checkbox.

[vasole]

The use_limits checkbox was introduced in PyMca 3.8 (May 2005) and as you can read it changed absolutely nothing for batch fitting:

• IMPORTANT: New fit region configuration handling. Some users get confused by the fact the fitting region is determined by the zoomed part of the data selection window. Now the user will have the choice. In the fit configuration window there is a check box related to the limits. In previous versions it was disabled. Now, when the check box is checked, the limits are determined by the fit configuration (user request), when the check box is not checked, the limits are determined by the zoomed part of the data selection window when starting the fit (behaviour till this release). For the batch fitting, I obviously force the fit to take the values from the config file.

[sergey-yaroslavtsev]

Not related to any batch. Here is just a single dummy data (channels vs channels from 0 to 1043). I fit with a simple constant (so the result is average value). If I select limits 0 1043 - no matter if the box is ticked or not I got result ~520. If I select limits 0 800 i got result 400.
I do not see what this box is doing in this case.

Related to batch. If I use StackSimpleFit and set configuration from file it applies this limits no matter if box is checked or not (same behavior as for single dataset). This create a problem when using "old" configuration to new data which is collected in different region.

[vasole]

I do not see what this box is doing in this case.

Zoom a different fitting region at the parent window. The one previous to reach the fit window. Then request a fit by clicking on the fit icon. In one case the fit region should be updated and in the other one not.

That is all what that checkbox is supposed to do.

[sergey-yaroslavtsev]

Ok, now I see what it is doing. However, it is not clear for the user at all, neither from the naming of check box neither how the limits are presented. I would consider to change at least the naming in this case.
At least It is misleading in such case: open fit window, untick the box -> expect not to have limits.
Because it is explicitly written "limit fitting region: from to" and unticked means "not" (see example above)

Another small issue remain - there is no way to disable the limits while setting configuration. Should be the possibility to set them to "None" (in other words "from -inf to inf") but GUI accepts only numbers.
If i create configuration in PyMca and then try to use this configuration to fit something else after loading configuration from file I always forced to change the limits manually to adapt limits to new data.

[vasole]

Ok, now I see what it is doing. However, it is not clear for the user at all

I disagree. It was requested by regular users of PyMca (not sporadic ones) and appreciated.

Another small issue remain - there is no way to disable the limits while setting configuration. Should be the possibility to set them to "None" (in other words "from -inf to inf") but GUI accepts only numbers.

From the GUI point of view, that is equivalent to having the checkbox not selected (the default behavior) and sending the whole spectrum. That is exactly what you achieve when plotting and going to the fit. Sending None is exactly that behavior.

The behavior is there and it is the default. I really do not understand that need to change something requested by users, used for more than 20 years without complains.

I think both of you are making things quite difficult when you are programmers and you can choose what to do when calling the code in a programmatic way. You are quite far from being the regular PyMca user.

If i create configuration in PyMca and then try to use this configuration to fit something else after loading configuration from file I always forced to change the limits manually to adapt limits to new data.

You just have to send the data after you have set the configuration. Not the other way around. The last thing a user wants is to have different results interactively and in a batch.

You have data to be fitted, if you load a configuration is because you want it to be used. Not the other way around.