ENH Speed up several functions.

RELEASE.rst

@@ -27,7 +27,11 @@ la 0.6
 - Upgrade numpydoc from 0.3.1 to 0.4 to support Sphinx 1.0.1
 - la.farray.ranking() and larry ranking method support `axis=None`
 - Generate C code with Cython 0.15.1 instead of Cython 0.11
-- larry.astype(), larry.push(), larry.__rdiv__() are faster
+
+**Faster**
+
+- larry methods: merge, nan_replace, push, cumsum, cumprod, astype, __rdiv__
+- Numpy array functions: geometric_mean, correlation, covMissing
 
 **Breakage from la 0.5**
 

la/deflarry.py

@@ -347,10 +347,10 @@ def cumsum(self, axis):
             raise ValueError, 'axis cannot be None'
         y = self.copy()
         idx = np.isnan(y.x)
-        y[idx] = 0
+        np.putmask(y.x, idx, 0)
         y.x.cumsum(axis, out=y.x)
         if idx.any():
-            y.x[idx] = np.nan
+            np.putmask(y.x, idx, np.nan)
         return y        
 
     def cumprod(self, axis): 
@@ -413,10 +413,10 @@ def cumprod(self, axis):
             raise ValueError, 'axis cannot be None'
         y = self.copy()
         idx = np.isnan(y.x)
-        y[idx] = 1
+        np.putmask(y.x, idx, 1)
         y.x.cumprod(axis, out=y.x)
         if idx.any():
-            y.x[idx] = np.nan
+            np.putmask(y.x, idx, np.nan)
         return y
 
     def clip(self, lo, hi):
@@ -1047,14 +1047,14 @@ def prod(self, axis=None):
         """
         y = self.copy()
         idx = np.isnan(y.x)
-        y.x[idx] = 1
+        np.putmask(y.x, idx, 1)
         y = y.__reduce(np.prod, axis=axis)
         idx = idx.all(axis)
         if idx.ndim == 0:
             if idx:
                 y = np.nan
-        else:        
-            y[idx] = np.nan
+        else:
+            np.putmask(y.x, idx, np.nan)
         return y 
 
     def mean(self, axis=None):
@@ -3062,9 +3062,9 @@ def quantile(self, q, axis=0):
         array([-1., -1.,  0.,  0.,  1.,  1.])
             
         """
-        y = self.copy()
-        y.x = quantile(y.x, q, axis=axis)       
-        return y
+        label = self.copylabel()
+        x = quantile(self.x, q, axis=axis)       
+        return larry(x, label, validate=False)
 
     # Group calc -------------------------------------------------------------  
 
@@ -3184,7 +3184,7 @@ def morph(self, label, axis):
                     miss = missing_marker(x)
                     if miss == NotImplemented:
                         x = x.astype(float)
-                        miss = missing_marker(x)      
+                        miss = missing_marker(x)
                     x[index] = miss      
             lab = self.copylabel()
             lab[axis] = list(label)
@@ -3900,7 +3900,7 @@ def nan_replace(self, replace_with=0):
                     
         """
         y = self.copy()
-        y.x[np.isnan(y.x)] = replace_with
+        np.putmask(y.x, np.isnan(y.x), replace_with)
         return y                                     
 
     # Size, shape, type ------------------------------------------------------

la/farray/misc.py

@@ -18,7 +18,7 @@ def geometric_mean(x, axis=-1, check_for_greater_than_zero=True):
         raise ValueError, msg
     x = x.copy()
     m = np.isnan(x)
-    x[m] = 1.0
+    np.putmask(x, m, 1.0)
     m = np.asarray(~m, np.float64)
     m = m.sum(axis)
     x = np.log(x).sum(axis)
@@ -30,7 +30,7 @@ def geometric_mean(x, axis=-1, check_for_greater_than_zero=True):
         if m == 0:
             idx = np.nan
     else:
-        idx[m == 0] = np.nan
+        np.putmask(idx, m==0, np.nan)
     x = np.multiply(x, idx)
     return x
 
@@ -97,8 +97,8 @@ def correlation(arr1, arr2, axis=None):
         else:
             x1 = arr1.copy()
             x2 = arr2.copy()
-            x1[mask] = np.nan
-            x2[mask] = np.nan 
+            np.putmask(x1, mask, np.nan)
+            np.putmask(x2, mask, np.nan)
             if axis == 0:
                 x1 = x1 - bn.nanmean(x1, axis)
                 x2 = x2 - bn.nanmean(x2, axis)              
@@ -152,7 +152,7 @@ def covMissing(R):
     
     """
     mask = np.isnan(R)
-    R[mask] = 0
+    np.putmask(R, mask, 0)
     mask = np.asarray(mask, np.float64)
     mask = 1 - mask # Change meaning of missing matrix to present matrix  
 

la/farray/normalize.py

@@ -120,7 +120,7 @@ def lastrank(x, axis=-1, decay=0.0):
         if not np.isfinite(x[indlast2]):
             r = np.nan
     else:
-        r[~np.isfinite(x[indlast2])] = np.nan
+        np.putmask(r, ~np.isfinite(x[indlast2]), np.nan)
     return r    
 
 def ranking(x, axis=0, norm='-1,1'):
@@ -195,7 +195,7 @@ def ranking(x, axis=0, norm='-1,1'):
     else:
         msg = "norm must be '-1,1', '0,N-1', or 'gaussian'."
         raise ValueError(msg)
-    idx[(countnotnan==1)*(~masknan)] = middle
+    np.putmask(idx, (countnotnan==1)*(~masknan), middle)
     return idx
 
 def push(x, n, axis=-1):
@@ -273,7 +273,7 @@ def quantile(x, q, axis=0):
         raise ValueError, 'q must be one or greater.'
     elif q == 1:
         y = np.zeros(x.shape)
-        y[np.isnan(x)] = np.nan
+        np.putmask(y, np.isnan(x), np.nan)
         return y
     if axis == None:
         if q > x.size: