refactor perf3

Author: h-vetinari

pandas/core/strings.py

@@ -36,30 +36,28 @@
 _shared_docs = dict()
 
 
-def str_cat_core(array, sep):
+def interleave_sep(all_cols, sep):
     '''
     Auxiliary function for :meth:`str.cat`
 
     Parameters
     ----------
-    array : ndarray
-        Array containing the vectors to be concatenated. These vectors must be
-        of object type and may not contain any nulls!
+    all_cols : list of numpy arrays
+        List of arrays to be concatenated with sep
     sep : string
         The separator string for concatenating the columns
 
     Returns
     -------
-    concatenated
-        the vector of concatenated results
+    list
+        The list of arrays interleaved with sep; to be fed to np.sum
     '''
     if sep == '':
-        return array.sum(axis=1)
-    else:
-        res = array[:, 0]
-        for i in range(1, array.shape[1]):
-            res += sep + array[:, i]
-        return res
+        # no need to add empty strings
+        return all_cols
+    result = [sep] * (2 * len(all_cols) - 1)
+    result[::2] = all_cols
+    return result
 
 
 def _na_map(f, arr, na_result=np.nan, dtype=object):
@@ -2098,11 +2096,13 @@ def cat(self, others=None, sep=None, na_rep=None, join=None):
 
         # concatenate Series/Index with itself if no "others"
         if others is None:
-            if na_rep is None:
-                data = data.dropna()
-            else:
-                data = data.fillna(na_rep)
-            return sep.join(data.values)
+            data = data.astype(object).values
+            mask = isna(data)
+            if mask.any():
+                if na_rep is None:
+                    return sep.join(data[~mask])
+                return sep.join(np.where(mask, na_rep, data))
+            return sep.join(data)
 
         try:
             # turn anything in "others" into lists of Series
@@ -2145,18 +2145,27 @@ def cat(self, others=None, sep=None, na_rep=None, join=None):
                             join=(join if join == 'inner' else 'outer'),
                             keys=range(len(others)), copy=False)
             data, others = data.align(others, join=join)
-            df = concat([data, others], axis=1, copy=False).astype('object')
-        else:
-            df = concat([data] + others, axis=1, copy=False).astype('object')
-
-        # calculate in numpy using str_cat_core; result is 1-dim np.ndarray
-        if na_rep is None:
-            mask = df.isna().values.any(axis=1)
-            result = np.full(len(data), fill_value=np.nan, dtype='object')
-            result[~mask] = str_cat_core(df.values[~mask], sep)
-        else:
-            df = df.fillna(na_rep)
-            result = str_cat_core(df.values, sep)
+            others = [others[x] for x in others]  # again list of Series
+
+        all_cols = [x.astype(object).values for x in [data] + others]
+        masks = np.array([isna(x) for x in all_cols])
+        union_mask = np.logical_or.reduce(masks, axis=0)
+
+        if na_rep is None and union_mask.any():
+            result = np.empty(len(data), dtype=object)
+            np.putmask(result, union_mask, np.nan)
+
+            not_masked = ~union_mask
+            all_cols = interleave_sep([x[not_masked] for x in all_cols], sep)
+
+            result[not_masked] = np.sum(all_cols, axis=0)
+        elif na_rep is not None and union_mask.any():
+            # fill NaNs
+            all_cols = [np.where(masks[i], na_rep, all_cols[i])
+                        for i in range(len(all_cols))]
+            result = np.sum(interleave_sep(all_cols, sep), axis=0)
+        else:  # no NaNs
+            result = np.sum(interleave_sep(all_cols, sep), axis=0)
 
         if isinstance(self._orig, Index):
             result = Index(result, name=self._orig.name)