Merge pull request #130 from paulmueller/develop

Develop

Author: Paul Müller

ChangeLog.txt

@@ -1,3 +1,11 @@
+0.9.0
+- Improve parameter display (#52, #114)
+- Display Chi2 on each page (#115)
+- The displayed Chi2-value for non-weighted fits is now
+  normalized to the expected values of the fit. The
+  documentation has been updated accordingly.
+- Add "All files" option in save dialogs (#97)
+- Improved plot export dialog (#99)
 0.8.9
 - Improved support for "ALV-7004" files (#104)
 - Increase resolution for image export

doc/PyCorrFit_doc_content.tex

@@ -696,11 +696,35 @@ \subsection{Fitting}
 
 \subsubsection{Weighted fitting}
 \label{sec:theor.weigh}
-In certain cases, it is useful to implement weights (standard deviation) $\sigma_i$ for the calculation of $\chi^2$. For example, very noisy parts of a correlation curve can falsify the resulting fit. In \textit{PyCorrFit}, weighting is implemented as follows:
+In certain cases, it is useful to perform weighted fitting with a known variance $\sigma_i^2$ at the data points $\tau_i$. In \textit{PyCorrFit}, weighted fitting is implemented as follows:
 \begin{equation}
 \chi^2_\mathrm{weighted} = \min_{\alpha_1, \dots, \alpha_k} \sum_{i=1}^n  \frac{\left[ G(\tau_i,\alpha_1, \dots, \alpha_k) - H(\tau_i) \right]^2}{\sigma_i^2}
 \end{equation}
-\textit{PyCorrFit} is able to calculate the weights $\sigma_i$ from the experimental data. The different approaches of this calculation of weights implemented in \textit{PyCorrFit} are explained in \hyref{Section}{sec:intro.graph}.
+
+Besides importing the variance alongside experimental data, \textit{PyCorrFit} is able to estimate the variance from the experimental data via several different approaches. A recommended approach is averaging over several curves. Other approaches such as estimation of the variance from spline fits or from the model function (see \hyref{Section}{sec:intro.graph}) cannot be considered unbiased. 
+Note that when performing global fits (see \hyref{Section}{sec:menub.tools.globa}), different types of weights for different correlation curves can strongly influence the result of the fit. Especially mixing curves with and without weights will most likely result in unphysical fits.
+
+\subsubsection{Displayed $\chi^2$ values}
+The displayed value of $\chi^2$ is defined by the type of the performed fit. This value is commonly normalized by the degrees of freedom $\nu = N - n - 1$, where $N$ is the number of observations (data points) and $n$ is the number of fitting parameters.
+\begin{itemize}
+
+\item \textbf{reduced expected sum of squares}: This value is used when there is no variance available for plot normalization.
+\begin{equation}
+\chi^2_\mathrm{red,exp} = \frac{1}{\nu}\sum_{i=1}^n  \frac{\left[ G(\tau_i,\alpha_\mathrm{min}) - H(\tau_i) \right]^2}{G(\tau_i,\alpha_\mathrm{min})}
+\end{equation}
+
+\item \textbf{reduced weighted sum of squares}: This  value is used when the fit was performed with variances $\sigma^2$.
+\begin{equation}
+\chi^2_\mathrm{red,weight} = \frac{1}{\nu}\sum_{i=1}^n  \frac{\left[ G(\tau_i,\alpha_\mathrm{min}) - H(\tau_i) \right]^2}{\sigma^2}
+\end{equation}
+
+\item \textbf{reduced global sum of squares}: This  value is used for global fits. The weights are computed identically to the situation with reduced weights, except that the variance $\sigma_\textrm{glob}^2$ may result in non-physical weighting (hence the emphasis on global).
+\begin{equation}
+\chi^2_\mathrm{red,weight} = \frac{1}{\nu}\sum_{i=1}^n  \frac{\left[ G(\tau_i,\alpha_\mathrm{min}) - H(\tau_i) \right]^2}{\sigma_\textrm{glob}^2}
+\end{equation}
+
+\end{itemize}
+
 
 \subsubsection{Algorithms}
 \label{sec:theor.alg}

pycorrfit/edclasses.py

@@ -23,35 +23,11 @@
     except:
         pass
 
-
-import numpy as np
 import sys
 import traceback
-from wx.lib.agw import floatspin        # Float numbers in spin fields
 import wx 
 
 
-class FloatSpin(floatspin.FloatSpin):
-    def __init__(self, parent, digits=10, increment=.01):
-        floatspin.FloatSpin.__init__(self, parent, digits=digits,
-                                     increment = increment)
-        self.Bind(wx.EVT_SPINCTRL, self.increment)
-        #self.Bind(wx.EVT_SPIN, self.increment)
-        #self.increment()
-
-
-    def increment(self, event=None):
-        # Find significant digit
-        # and use it as the new increment
-        x = self.GetValue()
-        if x == 0:
-            incre = 0.1
-        else:
-            digit = int(np.ceil(np.log10(abs(x)))) - 2
-            incre = 10**digit
-        self.SetIncrement(incre)
-
-
 class ChoicesDialog(wx.Dialog):
     def __init__(self, parent, dropdownlist, title, text):
         # parent is main frame
@@ -107,7 +83,7 @@ def save_figure(self, evt=None):
         Page = self.canvas.HACK_Page
         add = self.canvas.HACK_append
         dirname = parent.dirname
-        filename = Page.tabtitle.GetValue().strip()+Page.counter[:2]+add
+        filename = self.canvas.get_window_title().replace(" ", "_").lower()+add
         formats = fig.canvas.get_supported_filetypes()
     except:
         dirname = "."
@@ -129,12 +105,8 @@ def save_figure(self, evt=None):
     if dlg.ShowModal() == wx.ID_OK:
         wildcard = keys[dlg.GetFilterIndex()]
         filename = dlg.GetPath()
-        haswc = False
-        for key in keys:
-            if filename.lower().endswith("."+key) is True:
-                haswc = True
-        if haswc == False:
-            filename = filename+"."+wildcard
+        if not filename.endswith(wildcard):
+            filename += "."+wildcard
         dirname = dlg.GetDirectory()
         #savename = os.path.join(dirname, filename)
         savename = filename

pycorrfit/fcs_data_set.py

@@ -667,6 +667,7 @@ def __init__(self, correlations=[], global_fit=False,
         self.global_fit_variables = global_fit_variables
         self.verbose = verbose
         self.uselatex = uselatex
+        self.is_weighted_fit = False
 
         if not global_fit:
             # Fit each correlation separately
@@ -679,6 +680,7 @@ def __init__(self, correlations=[], global_fit=False,
                 # fit_bool: True for variable
                 self.fit_bool = corr.fit_parameters_variable.copy()
                 self.fit_parm = corr.fit_parameters.copy()
+                self.is_weighted_fit = corr.is_weighted_fit
                 self.fit_weights = Fit.compute_weights(corr,
                                                    verbose=verbose,
                                                    uselatex=uselatex)
@@ -705,6 +707,7 @@ def __init__(self, correlations=[], global_fit=False,
             varin = list()      # names of variable fitting parameters
             variv = list()      # values of variable fitting parameters
             varmap = list()     # list of indices of fitted parameters
+            self.is_weighted_fit = None
             for corr in self.correlations:
                 xtemp.append(corr.correlation_fit[:,0])
                 ytemp.append(corr.correlation_fit[:,1])
@@ -816,13 +819,15 @@ def get_fit_results(self, correlation):
         c = correlation
         d = {
              "chi2" : self.chi_squared,
+             "chi2 type" : self.chi_squared_type,
              "weighted fit" : c.is_weighted_fit,
              "fit algorithm" : c.fit_algorithm,
              "fit result" : c.fit_parameters.copy(),
              "fit parameters" : np.where(c.fit_parameters_variable)[0],
              "fit weights" : self.compute_weights(c)
              }
 
+        
         if c.is_weighted_fit:
             d["weighted fit type"] = c.fit_weight_type
             if isinstance(c.fit_weight_data, (int, float)):
@@ -837,14 +842,47 @@ def get_fit_results(self, correlation):
 
     @property
     def chi_squared(self):
-        """
-            Calculate Chi² for the current class.
+        """ Calculate displayed Chi²
+        
+            Calculate reduced Chi² for the current class.
         """
         # Calculate degrees of freedom
         dof = len(self.x) - np.sum(self.fit_bool) - 1
         # This is exactly what is minimized by the scalar minimizers
         chi2 = self.fit_function_scalar(self.fit_parm[self.fit_bool], self.x)
-        return chi2 / dof
+        if self.chi_squared_type == "reduced expected sum of squares":
+            fitted = self.func(self.fit_parm, self.x)
+            chi2 = np.sum((self.y-fitted)**2/np.abs(fitted)) / dof
+        elif self.chi_squared_type == "reduced weighted sum of squares":
+            fitted = self.func(self.fit_parm, self.x)
+            variance = self.fit_weights**2
+            chi2 = np.sum((self.y-fitted)**2/variance) / dof
+        elif self.chi_squared_type == "reduced global sum of squares":
+            fitted = self.func(self.fit_parm, self.x)
+            variance = self.fit_weights**2
+            chi2 = np.sum((self.y-fitted)**2/variance) / dof
+        return chi2
+
+
+    @property
+    def chi_squared_type(self):
+        """ The type of Chi² that currently applies.
+        
+        Returns
+        -------
+        "reduced" - if variance of data was used for fitting
+        "reduced Pearson" - if variance of data is not available
+        """
+        if self.is_weighted_fit is None:
+            # global fitting
+            return "reduced global sum of squares"
+        elif self.is_weighted_fit == True:
+            return "reduced weighted sum of squares"
+        elif self.is_weighted_fit == False:
+            return "reduced expected sum of squares"
+        else:
+            raise ValueError("Unknown weight type!")
+
 
     @staticmethod
     def compute_weights(correlation, verbose=0, uselatex=False):
@@ -1111,6 +1149,7 @@ def minimize(self):
         # Only allow physically correct parameters
         self.fit_parm = self.check_parms(self.fit_parm)
         # Write optimal parameters back to this class.
+        # Must be called after `self.fitparm = ...`
         chi = self.chi_squared
         # Compute error estimates for fit (Only "Lev-Mar")
         if self.fit_algorithm == "Lev-Mar":

pycorrfit/frontend.py

@@ -1454,8 +1454,8 @@ def OnSaveData(self,e=None):
         # Export CSV data
         filename = Page.tabtitle.GetValue().strip()+Page.counter[:2]+".csv"
         dlg = wx.FileDialog(self, "Save curve", self.dirname, filename, 
-              "Correlation with trace (*.csv)|*.csv;*.CSV"+\
-              "|Correlation only (*.csv)|*.csv;*.CSV",
+              "Correlation with trace (*.csv)|*.csv;*.*"+\
+              "|Correlation only (*.csv)|*.csv;*.*",
                wx.FD_SAVE|wx.FD_OVERWRITE_PROMPT)
         # user cannot do anything until he clicks "OK"
         if dlg.ShowModal() == wx.ID_OK:
@@ -1594,7 +1594,7 @@ def OnSaveSession(self,e=None):
         Infodict["Comments"]["Session"] = self.SessionComment
         # File dialog
         dlg = wx.FileDialog(self, "Save session file", self.dirname, "",
-                            "PyCorrFit session (*.pcfs)|*.pcfs",
+                            "PyCorrFit session (*.pcfs)|*.pcfs|All files (*.*)|*.*",
                             wx.FD_SAVE|wx.FD_OVERWRITE_PROMPT)
         if dlg.ShowModal() == wx.ID_OK:
             # Save everything