ENH: Speed up binariseCategorical with take by > 1000%

funpack/processing_functions_core.py

@@ -319,13 +319,18 @@ def binariseCategorical(data, minpres=None, take=None):
     # every column in it has the same dtype
     binshape = (len(data), len(uniq))
     if take is None: bindata = np.zeros(binshape, dtype=np.uint8)
-    else:            bindata = np.zeros(binshape, dtype=take.dtypes[0])
+    else:            bindata = np.full( binshape, np.nan,
+                                        dtype=take.dtypes[0])
 
     for i, v in enumerate(uniq):
         mask = data == v
-        if take is None: values = mask.any(axis=1)
-        else:            values = take.where(mask.values).apply(first, axis=1)
-        bindata[:, i] = values
+        if take is None:
+            bindata[:, i] = mask.any(axis=1)
+        else:
+            rowmask             = mask.any(axis=1)
+            idxs                = np.argmax(mask.values, axis=1)
+            values              = take.values[np.arange(take.shape[0]), idxs]
+            bindata[rowmask, i] = values[rowmask]
 
     return bindata, uniq