source:
sasview/src/sas/qtgui/Perspectives/Fitting/media/fitting_help.rst
@
2e5081b
Last change on this file since 2e5081b was 3b0b17e, checked in by wojciech, 6 years ago | |
---|---|
|
|
File size: 31.6 KB |
Fitting
Note
If some code blocks are not readable, expand the documentation window
Preparing to fit data
To fit some data you must first load some data, activate one or more data sets, send those data sets to fitting, and select a model to fit to each data set.
Instructions on how to load and activate data are in the section :ref:`Loading_data`.
SasView can fit data in one of three ways:
- in Single fit mode - individual data sets are fitted independently one-by-one
- in Simultaneous fit mode - multiple data sets are fitted simultaneously to the same model with/without constrained parameters (this might be useful, for example, if you have measured the same sample at different contrasts)
- in Batch fit mode - multiple data sets are fitted sequentially to the same model (this might be useful, for example, if you have performed a kinetic or time-resolved experiment and have lots of data sets!)
Selecting a model
The models in SasView are grouped into categories. By default these consist of:
- Cylinder - cylindrical shapes (disc, right cylinder, cylinder with endcaps etc)
- Ellipsoid - ellipsoidal shapes (oblate,prolate, core shell, etc)
- Parellelepiped - as the name implies
- Sphere - sheroidal shapes (sphere, core multishell, vesicle, etc)
- Lamellae - lamellar shapes (lamellar, core shell lamellar, stacked lamellar, etc)
- Shape-Independent - models describing structure in terms of density correlation functions, fractals, peaks, power laws, etc
- Paracrystal - semi ordered structures (bcc, fcc, etc)
- Structure Factor - S(Q) models
- Plugin Models - User-created (custom/non-library) Python models
Use the Category drop-down menu to chose a category of model, then select a model from the drop-down menu to the right. The "Show Plot" button on the bottom of the dialog will become active. If you click on it, a graph of the chosen model, calculated using default parameter values, will appear. The graph will update dynamically as the parameter values are changed.
Once you have selected a model you can read its help documentation by right clicking on the empty space in the parameter table.
Show 1D/2D
Models are normally fitted to 1D (ie, I(Q) vs Q) data sets, but some models in SasView can also be fitted to 2D (ie, I(Qx,Qy) vs Qx vs Qy) data sets.
NB: Magnetic scattering can only be fitted in SasView in 2D.
To activate 2D fitting mode, select the 2D view checkbox on the Fit Page. To return to 1D fitting model, de-select the same checkbox.
Model Functions
For a complete list of all the library models available in SasView, see the Model Documentation .
It is also possible to add your own models.
Adding your own Models
There are essentially three ways to generate new fitting models for SasView:
Using the SasView :ref:`New_Plugin_Model` helper dialog (best for beginners and/or relatively simple models)
By copying/editing an existing model (this can include models generated by the New Plugin Model* dialog) in the :ref:`Python_shell` or :ref:`Plugin_Editor` (suitable for all use cases)
By writing a model from scratch outside of SasView (only recommended for code monkeys!)
Please read the guidance on :ref:`Writing_a_Plugin` before proceeding.
To be found by SasView your model must reside in the *~\.sasview\plugin_models* folder.
Plugin Model Operations
From the Fitting option in the menu bar, select one of the options:
- Add Custom Model - to create a plugin model template with a helper dialog
- Edit Custom Model - to edit a plugin model in an editor window
- Manage Custom Models - to open a custom model manager allowing for a number of actions to be taken on custom models: listing, adding, deleteing, duplicating, editing
- Add/Multiply Models - to create a plugin model by summing/multiplying existing models in the model library
Add Custom Model
Relatively straightforward models can be programmed directly from the SasView GUI using the Plugin Definition Function.
When using this feature, be aware that even if your code has errors, including syntax errors, a model file is still generated. When you then correct the errors and click 'Apply' again to re-compile you will get an error informing you that the model already exists if the 'Overwrite' box is not checked. In this case you will need to supply a new model function name. By default the 'Overwrite' box is checked.
Also note that the 'Fit Parameters' have been split into two sections: those which can be polydisperse (shape and orientation parameters) and those which are not (eg, scattering length densities).
A model file generated by this option can be viewed and further modified using the :ref:`Model_Editor`.
SasView version 4.2 made it possible to specify whether a plugin created with the New Plugin Model dialog is actually a form factor P(Q) or a structure factor S(Q). To do this, simply add one or other of the following lines under the import statements.
For a form factor:
form_factor = True
or for a structure factor:
structure_factor = True
If the plugin is a structure factor it is also necessary to add two variables to the parameter list:
parameters = [ ['radius_effective', '', 1, [0.0, numpy.inf], 'volume', ''], ['volfraction', '', 1, [0.0, 1.0], '', ''], [...],
and to the declarations of the functions Iq and Iqxy::
def Iq(x , radius_effective, volfraction, ...): def Iqxy(x, y, radius_effective, volfraction, ...):
Such a plugin should then be available in the S(Q) drop-down box on a FitPage (once a P(Q) model has been selected).
Model Editor
Selecting "Edit Custom Model" option opens the editor window.
Initially, the editor is empty. A custom model can be loaded by clicking on the Load plugin... button and choosing one of the existing custom plugins.
Once the model is loaded, it can be edited and saved with Save button. Saving the model will perform the validation and only when the model is correct it will be saved to a file. Successful model check is indicated by a SasView status bar message.
When Cancel is clicked, any changes to the model are discarded and the window is closed.
For details of the SasView plugin model format see :ref:`Writing_a_Plugin` .
To use the model, go to the relevant Fit Page, select the Plugin Models category and then select the model from the drop-down menu.
Plugin Manager
Selecting the Manage Custom Models option shows a list of all the plugin models in the plugin model folder, on Windows this is
C:\Users\{username}\.sasview\plugin_models
You can add, edit, duplicate and delete these models using buttons on the right side of the list.
Add a model
Clicking the "Add" button opens the Model Editor window, allowing you to create a new plugin as described above.
Duplicate a model
Clicking the "Duplicate" button will create a copy of the selected model(s). Naming of the duplicate follows the standard, with added * (n)* to the plugin model name, with n being the first unused yet integer.
Edit a model
When a single model is selected, clicking this button will open the Advanced Model Editor allowing you to edit the Python code of the model. If no models or multiple models are selected, the Edit button is disabled.
Delete Plugin Models
Simply highlight the plugin model(s) to be removed and click on the "Delete" button. The operation is final.
NB: Models shipped with SasView cannot be removed in this way.
Add/Multiply Models
This option creates a custom Plugin Model of the form:
Plugin Model = scale_factor * {(scale_1 * model_1) +/- (scale_2 * model_2)} + background
or:
Plugin Model = scale_factor * (model1 * model2) + background
In the Add/Multiply Models give the new model a function name and brief description (to appear under the Details button on the FitPage). Then select two existing models, as model_1 and model_2, and the required operator, '+' or '*' between them. Finally, click the Apply button to generate and test the model and then click Close.
SasView version 4.2 introduced a much simplified and more extensible structure for plugin models generated through the Add/Multiply Models editor. For example, the code for a combination of a sphere model with a power law model now looks like this:
from sasmodels.core import load_model_info from sasmodels.sasview_model import make_model_from_info model_info = load_model_info('sphere+power_law') model_info.name = 'MyPluginModel' model_info.description = 'sphere + power_law' Model = make_model_from_info(model_info)
To change the models or operators contributing to this plugin it is only necessary to edit the string in the brackets after load_model_info, though it would also be a good idea to update the model name and description too!!!
The model specification string can handle multiple models and combinations of operators (+ or ) which are processed according to normal conventions. Thus 'model1+model2*model3' would be valid and would multiply model2 by model3 before adding model1. In this example, parameters in the *FitPage would be prefixed A (for model2), B (for model3) and C (for model1). Whilst this might appear a little confusing, unless you were creating a plugin model from multiple instances of the same model the parameter assignments ought to be obvious when you load the plugin.
If you need to include another plugin model in the model specification string, just prefix the name of that model with custom. For instance:
sphere+custom.MyPluginModel
To create a P(Q)*S(Q) model use the @ symbol instead of * like this:
sphere@hardsphere
This streamlined approach to building complex plugin models from existing library models, or models available on the Model Marketplace, also permits the creation of P(Q)*S(Q) plugin models, something that was not possible in earlier versions of SasView.
Fit Algorithms
It is possible to specify which optimiser SasView should use to fit the data, and to modify some of the configurational parameters for each optimiser.
From Fitting in the menu bar select Fit Algorithms, then select one of the following optimisers:
- DREAM
- Levenberg-Marquardt
- Quasi-Newton BFGS
- Differential Evolution
- Nelder-Mead Simplex
The DREAM optimiser is the most sophisticated, but may not necessarily be the best option for fitting simple models. If uncertain, try the Levenberg-Marquardt optimiser initially.
These optimisers form the Bumps package written by P Kienzle. For more information on each optimiser, see the :ref:`Fitting_Documentation`.
Fitting Limits
By default, SasView will attempt to model fit the full range of the data; ie, across all Q values. If necessary, however, it is possible to specify only a sub-region of the data for fitting.
In a FitPage or BatchPage change the tab to Fit Options and then change the Q values in the Min and/or Max text boxes.
To return to including all data in the fit, click the Reset button.
Shortcuts ---------Copy/Paste Parameters ^^^^^^^^^^^^^^^^^^^^^It is possible to copy the parameters from one Fit Page and to paste them into another Fit Page using the same model.
To copy parameters, either:
- Select Edit -> Copy Params from the menu bar, or
- Use Ctrl(Cmd on Mac) + Left Mouse Click on the Fit Page.
To paste parameters, either:
- Select Edit -> Paste Params from the menu bar, or
- Use Ctrl(Cmd on Mac) + Shift + Left-click on the Fit Page.
If either operation is successful a message will appear in the info line at the bottom of the SasView window.
Bookmark ^^^^^^^^To Bookmark a Fit Page either:
- Select a Fit Page and then click on Bookmark in the tool bar, or
- Right-click and select the Bookmark in the popup menu.
Status Bar & Log Explorer
The status bar is located at the bottom of the SasView window and displays messages, warnings and errors.
The bottom part of the SasView application window contains the Log Explorer. The Log Explorer displays available message history and run-time traceback information.
Single Fit Mode
NB: Before proceeding, ensure that the Batch mode checkbox at the bottom of the Data Explorer is unchecked (see the section :ref:`Loading_data` ).
This mode fits one data set.
When data is sent to the fitting, the Fit Page will show the dataset name.
Clicking on the Show Plot will cause the data can be plotted in a graph window as markers.
If a graph does not appear, or a graph window appears but is empty, then the data has not loaded correctly. Check to see if there is a message in the :ref:`Status_Bar` or in the Console window.
Assuming the data has loaded correctly, when a model is selected a blue model calculation (or what SasView calls a 'Theory') line will appear in the earlier graph window, and a second graph window will appear displaying the residuals (the difference between the experimental data and the theory) at the same X-data values. See :ref:`Assessing_Fit_Quality`.
The objective of model-fitting is to find a physically-plausible model, and set of model parameters, that generate a theory that reproduces the experimental data and gives residual values as close to zero as possible.
Change the default values of the model parameters by hand until the theory line starts to represent the experimental data. Then uncheck the tick boxes alongside all parameters except the 'background' and the 'scale'. Click the Fit button. SasView will optimise the values of the 'background' and 'scale' and also display the corresponding uncertainties on the optimised values.
NB: If no uncertainty is shown it generally means that the model is not very dependent on the corresponding parameter (or that one or more parameters are 'correlated').
In the bottom right corner of the Fit Page is a box displaying the normalised value of the statistical $chi^2$ parameter returned by the optimiser.
Now check the box for another model parameter and click Fit again. Repeat this process until most or all parameters are checked and have been optimised. As the fit of the theory to the experimental data improves the value of 'chi2/Npts' will decrease. A good model fit should easily produce values of 'chi2/Npts' that are close to one, and certainly <100. See :ref:`Assessing_Fit_Quality`.
SasView has a number of different optimisers (see the section :ref:`Fitting_Options`). The DREAM optimiser is the most sophisticated, but may not necessarily be the best option for fitting simple models. If uncertain, try the Levenberg-Marquardt optimiser initially.
Simultaneous Fit Mode
NB: Before proceeding, ensure that the Single Mode radio button at the bottom of the Data Explorer is checked (see the section :ref:`Loading_data` ).
This mode is an extension of the :ref:`Single_Fit_Mode` that fits two or more data sets to the same model simultaneously. If necessary it is possible to constrain fit parameters between data sets (eg, to fix a background level, or radius, etc).
If the data to be fit are in multiple files, load each file, then select each file in the Data Explorer, and Send To Fitting. If multiple data sets are in one file, load that file, Unselect All Data, select just those data sets to be fitted, and Send To Fitting. Either way, the result should be that for n data sets you have 2n graphs (n of the data and model fit, and n of the resulting residuals). But it may be helpful to minimise the residuals plots for clarity. Also see :ref:`Assessing_Fit_Quality`.
NB: If you need to use a custom Plugin Model, you must ensure that model is available first (see :ref:`Adding_your_own_models` ).
Method
Now go to each FitPage in turn and:
Select the required category and model;
Unselect all the model parameters;
Enter some starting guesses for the parameters;
Enter any parameter limits (recommended);
Select which parameters will refine (selecting all is generally a bad idea...);
When done, select Constrained or Simultaneous Fit under Fitting in the menu bar.
In the Const & Simul Fit page that appears, select which data sets are to be simultaneously fitted (this will probably be all of them or you would not have loaded them in the first place!).
To tie parameters between the data sets with constraints, select the data sets and right click. From the menu choose Mutual constraint of parameters in selected models
When ready, click the Fit button on the Const & Simul Fit page, NOT the Fit button on the individual FitPage's.
Simultaneous Fits without Constraints
The results of the model-fitting will be returned to each of the individual FitPage's.
Note that the chi2/Npts value returned is the SUM of the chi2/Npts of each fit. To see the chi2/Npts value for a specific FitPage, click the Compute button at the bottom of that FitPage to recalculate. Also see :ref:`Assessing_Fit_Quality`.
Simultaneous Fits with Constraints
In the Const. & Simul. Fit page select two data sets which are to be constrained. Right clicking in the Source choice for simultaneous fitting area will bring up the context menu:
Here you can choose datasets for fitting and define constraints between parameters in both datasets.
Selecting this option will bring up the Complex Constraint dialog.
Constraints will generally be of the form
Mi Parameter1 = Mj.Parameter1
however the text box after the '=' sign can be used to adjust this relationship; for example
Mi Parameter1 = scalar * Mj.Parameter1
A 'free-form' constraint box is also provided.
Many constraints can be entered for a single fit.
The results of the model-fitting will be returned to each of the individual FitPage's.
Note that the chi2/Npts value returned is the SUM of the chi2/Npts of each fit. To see the chi2/Npts value for a specific FitPage, click the Compute button at the bottom of that FitPage to recalculate. Also see :ref:`Assessing_Fit_Quality`.
Batch Fit Mode
NB: Before proceeding, ensure that the Single Mode radio button at the bottom of the Data Explorer is checked (see the section :ref:`Loading_data` ). The Batch Mode button will be used later on!
This mode sequentially fits two or more data sets to the same model. Unlike in simultaneous fitting, in batch fitting it is not possible to constrain fit parameters between data sets.
If the data to be fit are in multiple files, load each file in the Data Explorer. If multiple data sets are in one file, load just that file. Unselect All Data, then select a single initial data set to be fitted. Fit that selected data set as described above under :ref:`Single_Fit_Mode`.
NB: If you need to use a custom Plugin Model, you must ensure that model is available first (see :ref:`Adding_your_own_models` ).
Method
Now Select All Data in the Data Explorer, check the Batch Mode radio button at the bottom of that panel and Send To Fitting. A BatchPage will be created.
NB: The Batch Page can also be created by checking the Batch Mode radio button and selecting New Fit Page under Fitting in the menu bar.
Using the drop-down menus in the BatchPage, now set up the same data set with the same model that you just fitted in single fit mode. A quick way to set the model parameter values is to just copy them from the earlier Single Fit. To do this, go back to the Single Fit FitPage, select Copy Params under Edit in the menu bar, then go back to the BatchPage and Paste Params.
When ready, use the Fit button on the BatchPage to perform the fitting, NOT the Fit button on the individual FitPage's.
Unlike in single fit mode, the results of batch fits are not returned to the BatchPage. Instead, a spreadsheet-like :ref:`Grid_Window` will appear.
If you want to visually check a graph of a particular fit, click on the name of a Data set in the Grid Window and then click the View Fits button. The data and the model fit will be displayed. If you select mutliple data sets they will all appear on one graph.
NB: In theory, returning to the BatchPage and changing the name of the I(Q) data source should also work, but at the moment whilst this does change the data set displayed it always superimposes the 'theory' corresponding to the starting parameters.
If you select a 'Chi2' value and click the View Fits button a graph of the residuals for that data set is displayed. Again, if you select multiple 'Chi2' values then all the residuals data will appear on one graph. Also see :ref:`Assessing_Fit_Quality`.
Chain Fitting
By default, the same parameter values copied from the initial single fit into the BatchPage will be used as the starting parameters for all batch fits. It is, however, possible to get SasView to use the results of a fit to a preceding data set as the starting parameters for the next fit in the sequence. This variation of batch fitting is called Chain Fitting, and will considerably speed up model-fitting if you have lots of very similar data sets where a few parameters are gradually changing. Do not use chain fitting on disparate data sets.
To use chain fitting, select Chain Fitting under Fitting in the menu bar. It toggles on/off, so selecting it again will switch back to normal batch fitting.
Grid Window
The Grid Window provides an easy way to view the results from batch fitting. It will be displayed automatically when a batch fit completes, but may be opened at any time by selecting Show Grid Window under View in the menu bar.
If there is an existing Grid Window and another batch fit is performed,* an additional 'Table' tab will be added to the Grid Window.
The parameter values in the currently selected table of the Grid Window can be output to a CSV file by choosing Save As under File in the (Grid Window) menu bar. The default filename includes the date and time that the batch fit was performed.
Saved CSV files can be reloaded by choosing Open under File in the Grid Window menu bar. The loaded parameters will appear in a new table tab.
NB: Saving the Grid Window does not save any experimental data, residuals or actual model fits. Consequently if you reload a saved CSV file the ability to View Fits will be lost.
Parameter Plots
Any row (dataset) can be plotted by selecting it and either right-clicking and choosing Plot selected fits menu item or by clicking on the Plot button.
Combined Batch Fit Mode
The purpose of the Combined Batch Fit is to allow running two or more batch fits in sequence without overwriting the output table of results. This may be of interest for example if one is fitting a series of data sets where there is a shape change occurring in the series that requires changing the model part way through the series; for example a sphere to rod transition. Indeed the regular batch mode does not allow for multiple models and requires all the files in the series to be fit with single model and set of parameters. While it is of course possible to just run part of the series as a batch fit using model one followed by running another batch fit on the rest of the series with model two (and/or model three etc), doing so will overwrite the table of outputs from the previous batch fit(s). This may not be desirable if one is interested in comparing the parameters: for example the sphere radius of set one and the cylinder radius of set two.
Method
In order to use the Combined Batch Fit, first load all the data needed as described in :ref:`Loading_data`. Next start up two or more BatchPage fits following the instructions in :ref:`Batch_Fit_Mode` but DO NOT PRESS FIT.
When done, select Constrained or Simultaneous Fit under Fitting in the menu bar.
In the Const & Simul Fit page that appears, choose Batch fits radio button and select which data sets are to be fitted.
Once all are selected, click the Fit button on the Const. Simult. Fitting to run each batch fit in sequence
The batch table will then pop up at the end as for the case of the simple Batch Fitting with the following caveats:
Note
The order matters. The parameters in the table will be taken from the model used in the first BatchPage of the list. Any parameters from the second and later BatchPage s that have the same name as a parameter in the first will show up allowing for plotting of that parameter across the models. The other parameters will not be available in the grid.
Note
a corralary of the above is that currently models created as a sum|multiply model will not work as desired because the generated model parameters have a p#_ appended to the beginning and thus radius and p1_radius will not be recognized as the same parameter.
Note
This help document was last changed by Piotr Rozyczko, 18 May 2018