PERF: significant speedup in sparse init and ops by using numpy in check_integrity

Author: qwhelan

doc/source/whatsnew/v0.24.1.rst

@@ -21,6 +21,12 @@ including other versions of pandas.
 Enhancements
 ^^^^^^^^^^^^
 
+.. _whatsnew_0241.performance:
+
+Performance Improvements
+~~~~~~~~~~~~~~~~~~~~~~~~
+ - Significant speedup in `SparseArray` initialization that benefits most operations, fixing performance regression introduced in v0.20.0
+
 
 .. _whatsnew_0241.bug_fixes:
 

pandas/_libs/sparse.pyx

@@ -72,9 +72,6 @@ cdef class IntIndex(SparseIndex):
         A ValueError is raised if any of these conditions is violated.
         """
 
-        cdef:
-            int32_t index, prev = -1
-
         if self.npoints > self.length:
             msg = ("Too many indices. Expected "
                    "{exp} but found {act}").format(
@@ -86,17 +83,15 @@ cdef class IntIndex(SparseIndex):
         if self.npoints == 0:
             return
 
-        if min(self.indices) < 0:
+        if self.indices.min() < 0:
             raise ValueError("No index can be less than zero")
 
-        if max(self.indices) >= self.length:
+        if self.indices.max() >= self.length:
             raise ValueError("All indices must be less than the length")
 
-        for index in self.indices:
-            if prev != -1 and index <= prev:
-                raise ValueError("Indices must be strictly increasing")
-
-            prev = index
+        monotonic = np.all(self.indices[:-1] < self.indices[1:])
+        if not monotonic:
+            raise ValueError("Indices must be strictly increasing")
 
     def equals(self, other):
         if not isinstance(other, IntIndex):