fix?

ostap/stats/gof.py

@@ -129,10 +129,135 @@ def nEff ( weights ) :
 #  ds = ... # data set as structured array
 #  dsn = normalize ( ds ) 
 #  @endcode
+def normalize2 ( datasets , weight = () , first = True ) :
+    """ Get the `normalized' input datasets
+    All floating felds  are calculated as
+    
+    x = (x - <x>)/sigma
+    
+    - <x> is a mean value 
+    - is a standard deviation.
+    
+    - If several datasets are specified, all floating names must be the same
+    and the mean and sigma are either taken either from the first dataset,
+    if `first=True` or as combined through all datasets, otherwise  
+    
+    - If `weight` is specified, this floating column is concidered
+    as the weight
+    
+    - attention Only the floating point columns are transformed! 
+    - attention Input datasets are expected to be numpy structured arrays 
+    """
+
+    nd = len ( datasets ) 
+    if not weight                             : weight = nd * [ ''     ]
+    elif isinstance ( weight , string_types ) : weight = nd * [ weight ]
+
+    assert ( len ( weight ) == nd ) and \
+           all ( ( not w ) or isinstance ( w , string_types ) for w in weight ) , \
+           'Invalid specification of weight!'
+
+    weight = list ( weight )
+    for i , w in enumerate ( weight ) :
+        if not w : weight [ i ] = '' 
+        weight = tuple ( weight )
 
-if (3,0) <= sys.version_info :
+    ds     = datasets [ 0 ]
+    others = datasets [ 1: ]
+
+    ## collect the floating columns 
+    columns = []
+    w0      = weight [ 0 ] 
+    for n,t in ds.dtype.fields.items () :
+        if t[0] in _np_floats  and n != w0 : columns.append ( n ) 
+
+    vmeans  = [] 
+    for i , c in enumerate ( columns ) :
+        mean, var    = mean_var ( ds [ c ] , None if not w0 else ds [ w0 ] )
+        vmeans.append ( VE ( mean , var ) )
+
+    ## Number of events/effective entries 
+    nevents = 1.0 * ds.shape [ 0 ] if not w0 else nEff ( ds [ w0 ] )
 
-    def normalize ( ds , *others , weight = () , first = True ) :
+    if not first and others : 
+        nevents = ds.shape[0] 
+        for k , dd in enumerate ( others ) :
+
+            wk = weight [ k + 1 ] 
+            nn = 1.0 * dd.shape [ 0 ] if not wk else nEff ( dd [ wk ] )                
+
+            for i , c in enumerate ( columns ) :
+
+                mean, var = mean_var ( dd [ c ] , None if not wk else dd [ wk ] )                    
+                vv = VE ( mean , var ) 
+                vmean  [ i ] = Ostap.Math.two_samples ( vmean [ i ] , nevents , vv , nn )
+
+            nevents += nn
+
+    result = []  
+    for d in datasets :
+
+        nds = d.copy ()
+        for ic , c in enumerate ( columns ) :
+            vv        = vmeans [ ic ]
+            mean      = vv.value ()
+            sigma     = vv.error ()                 
+            a         = nds [ c ]
+            nds [ c ] =  ( a - mean ) / sigma
+
+        result.append ( nds )
+
+    return tuple ( result ) 
+
+
+
+# =============================================================================
+if (3,0) <= sys.version_info : 
+    # =========================================================================
+    ## Get the "normalized" input datasets
+    #  All floating felds  are calculated as
+    #  \f[ x = \frac{x - \left\langle x \right\rangle}{\sigma} \f]
+    #  where \f$ \left\langle x \right\rangle\f$ is mena value
+    #  and \f$ \sigma \f$ is a standard deviation.
+    # 
+    #  @code
+    #  ds = ... # data set as structured array
+    #  dsn = normalize ( ds ) 
+    #  @endcode
+    #
+    #  - If several datasets are specified, all floating names must be the same
+    #  and the mean and sigma are either taken either from the first dataset,
+    #  if <code>first=True</code> or as combined through all datasets otherwise 
+    #
+    #  @code
+    #  ds1 = ... # data set as structured array
+    #  ds2 = ... # data set as structured array
+    #  ds3 = ... # data set as structured array
+    #  ds1n, ds2n, ds3n  = normalize ( ds1 , ds2 , ds3 , first = True ) 
+    #  @endcode
+    #
+    #  - If <code>weight</code> is specified, this floating column is considered
+    #  as the weight
+    #
+    #  @code
+    #  ds = ... # data set as structured array with weight 
+    #  dsn = normalize ( ds , weight = 'weight' ) 
+    #  @endcode
+    #
+    #  @code
+    #  ds1 = ... # data set as structured array without weight 
+    #  ds2 = ... # data set as structured array with weight 
+    #  ds1n , ds2n = normalize ( ds1 , ds2 , weight = ( None , 'weight'  ) ) 
+    #  @endcode
+    #
+    #  @attention Only the floating point columns are transformed! 
+    #  @attention Input datasets are expected to be numpy structured arrays
+    #
+    #  @code
+    #  ds = ... # data set as structured array
+    #  dsn = normalize ( ds ) 
+    #  @endcode
+    def normalize ( ds , *others , weight = () , first = True ) :        
         """ Get the `normalized' input datasets
         All floating felds  are calculated as
         
@@ -152,68 +277,10 @@ def normalize ( ds , *others , weight = () , first = True ) :
         - attention Input datasets are expected to be numpy structured arrays 
         """
 
-        nd = len ( others ) + 1 
-        if not weight                             : weight = nd * [ ''     ]
-        elif isinstance ( weight , string_types ) : weight = nd * [ weight ]
-
-        assert ( len ( weight ) == nd ) and \
-               all ( ( not w ) or isinstance ( w , string_types ) for w in weight ) , \
-               'Invalid specification of weight!'
-
-        weight = list ( weight )
-        for i , w in enumerate ( weight ) :
-            if not w : weight [ i ] = '' 
-        weight = tuple ( weight )
-
-        data    = ( ds , ) + others  
-        result  = []  
-
-        ## collect the floating columns 
-        columns = []
-        w0      = weight [ 0 ] 
-        for n,t in ds.dtype.fields.items () :
-            if t[0] in _np_floats  and n != w0 : columns.append ( n ) 
-
-
-        vmeans  = [] 
-        for i , c in enumerate ( columns ) :
-            mean, var    = mean_var ( ds [ c ] , None if not w0 else ds [ w0 ] )
-            vmeans.append ( VE ( mean , var ) )
-
-        ## Number of events/effective entries 
-        nevents = 1.0 * ds.shape [ 0 ] if not w0 else nEff ( ds [ w0 ] )
-
-        if not first and others : 
-            nevents = ds.shape[0] 
-            for k , dd in enumerate ( others ) :
-
-                wk = weight [ k + 1 ] 
-                nn = 1.0 * dd.shape [ 0 ] if not wk else nEff ( dd [ wk ] )                
-
-                for i , c in enumerate ( columns ) :
-
-                    mean, var = mean_var ( dd [ c ] , None if not wk else dd [ wk ] )                    
-                    vv = VE ( mean , var ) 
-                    vmean  [ i ] = Ostap.Math.two_samples ( vmean [ i ] , nevents , vv , nn )
-
-                nevents += nn
-
-        result = []  
-        for d in ( ds , *others ) :
-
-            nds = d.copy ()
-            for ic , c in enumerate ( columns ) :
-                vv        = vmeans [ ic ]
-                mean      = vv.value ()
-                sigma     = vv.error ()                 
-                a         = nds [  c ]
-                nds [ c ] =  ( a - mean ) / sigma
-
-            result.append ( nds )
-
-        return result [ 0 ] if not others else tuple ( result ) 
-
+        result = normalize2 ( ( ds , *others ) , weight = weight , first = first )  
 
+        return result [ 0 ] if not others else resut 
+
 # =============================================================================
 if '__main__' == __name__ :