BUG: fix rolling_max/min for small inputs and large windows. Add a check that the min_period <= window size. Fixes pandas-dev#1897.

Author: erg

pandas/src/moments.pyx

@@ -9,6 +9,18 @@
 #
 # -
 
+def _check_minp(win, minp, N):
+    if minp > win:
+        raise ValueError('min_periods %d must be <= window %d'
+                        % (minp, win))
+    elif minp > N:
+        minp = N + 1
+    elif minp == 0:
+        minp = 1
+    elif minp < 0:
+        raise ValueError('min_periods must be >= 0')
+    return minp
+
 # original C implementation by N. Devillard.
 # This code in public domain.
 # Function :   kth_smallest()
@@ -24,7 +36,6 @@
 # Physical description: 366 p.
 #               Series: Prentice-Hall Series in Automatic Computation
 
-
 def kth_smallest(ndarray[double_t] a, Py_ssize_t k):
     cdef:
         Py_ssize_t i,j,l,m,n
@@ -149,7 +160,7 @@ def roll_sum(ndarray[double_t] input, int win, int minp):
 
     cdef ndarray[double_t] output = np.empty(N, dtype=float)
 
-    minp = _check_minp(minp, N)
+    minp = _check_minp(win, minp, N)
 
     for i from 0 <= i < minp - 1:
         val = input[i]
@@ -192,7 +203,7 @@ def roll_mean(ndarray[double_t] input,
 
     cdef ndarray[double_t] output = np.empty(N, dtype=float)
 
-    minp = _check_minp(minp, N)
+    minp = _check_minp(win, minp, N)
 
     for i from 0 <= i < minp - 1:
         val = input[i]
@@ -335,23 +346,14 @@ def nancorr(ndarray[float64_t, ndim=2] mat, cov=False):
 #----------------------------------------------------------------------
 # Rolling variance
 
-def _check_minp(minp, N):
-    if minp > N:
-        minp = N + 1
-    elif minp == 0:
-        minp = 1
-    elif minp < 0:
-        raise ValueError('min_periods must be >= 0')
-    return minp
-
 def roll_var(ndarray[double_t] input, int win, int minp, int ddof=1):
     cdef double val, prev, sum_x = 0, sum_xx = 0, nobs = 0
     cdef Py_ssize_t i
     cdef Py_ssize_t N = len(input)
 
     cdef ndarray[double_t] output = np.empty(N, dtype=float)
 
-    minp = _check_minp(minp, N)
+    minp = _check_minp(win, minp, N)
 
     for i from 0 <= i < minp - 1:
         val = input[i]
@@ -400,7 +402,7 @@ def roll_skew(ndarray[double_t] input, int win, int minp):
     # 3 components of the skewness equation
     cdef double A, B, C, R
 
-    minp = _check_minp(minp, N)
+    minp = _check_minp(win, minp, N)
 
     for i from 0 <= i < minp - 1:
         val = input[i]
@@ -462,7 +464,7 @@ def roll_kurt(ndarray[double_t] input,
     # 5 components of the kurtosis equation
     cdef double A, B, C, D, R, K
 
-    minp = _check_minp(minp, N)
+    minp = _check_minp(win, minp, N)
 
     for i from 0 <= i < minp - 1:
         val = input[i]
@@ -533,7 +535,7 @@ cdef _roll_skiplist_op(ndarray arg, int win, int minp, skiplist_f op):
 
     skiplist = IndexableSkiplist(win)
 
-    minp = _check_minp(minp, N)
+    minp = _check_minp(win, minp, N)
 
     for i from 0 <= i < minp - 1:
         val = input[i]
@@ -578,7 +580,7 @@ def roll_median_c(ndarray[float64_t] arg, int win, int minp):
 
     sl = skiplist_init(win)
 
-    minp = _check_minp(minp, N)
+    minp = _check_minp(win, minp, N)
 
     for i from 0 <= i < minp - 1:
         val = arg[i]
@@ -672,11 +674,17 @@ def roll_max2(ndarray[float64_t] a, int window, int minp):
     cdef np.ndarray[np.float64_t, ndim=1] y = PyArray_EMPTY(1, dims,
 		NPY_float64, 0)
 
-    minp = _check_minp(minp, n0)
+    if window < 1:
+        raise ValueError('Invalid window size %d'
+                         % (window))
 
-    if (window < 1) or (window > n0):
-        raise ValueError('Invalid window size %d for len %d array'
-                         % (window, n0))
+    if minp > window:
+        raise ValueError('Invalid min_periods size %d greater than window %d'
+                        % (minp, window))
+
+    minp = _check_minp(window, minp, n0)
+
+    window = min(window, n0)
 
     ring = <pairs*>stdlib.malloc(window * sizeof(pairs))
     end = ring + window
@@ -766,11 +774,18 @@ def roll_min2(np.ndarray[np.float64_t, ndim=1] a, int window, int minp):
     cdef np.npy_intp *dims = [n0]
     cdef np.ndarray[np.float64_t, ndim=1] y = PyArray_EMPTY(1, dims,
 		NPY_float64, 0)
-    if (window < 1) or (window > n0):
-        raise ValueError('Invalid window size %d for len %d array'
-                         % (window, n0))
 
-    minp = _check_minp(minp, n0)
+    if window < 1:
+        raise ValueError('Invalid window size %d'
+                         % (window))
+
+    if minp > window:
+        raise ValueError('Invalid min_periods size %d greater than window %d'
+                        % (minp, window))
+
+    window = min(window, n0)
+
+    minp = _check_minp(window, minp, n0)
 
     ring = <pairs*>stdlib.malloc(window * sizeof(pairs))
     end = ring + window
@@ -843,7 +858,7 @@ def roll_quantile(ndarray[float64_t, cast=True] input, int win,
 
     skiplist = IndexableSkiplist(win)
 
-    minp = _check_minp(minp, N)
+    minp = _check_minp(win, minp, N)
 
     for i from 0 <= i < minp - 1:
         val = input[i]
@@ -892,7 +907,7 @@ def roll_generic(ndarray[float64_t, cast=True] input, int win,
     if n == 0:
         return input
 
-    minp = _check_minp(minp, n)
+    minp = _check_minp(win, minp, n)
     output = np.empty(n, dtype=float)
     counts = roll_sum(np.isfinite(input).astype(float), win, minp)
 

pandas/stats/tests/test_moments.py

@@ -50,9 +50,21 @@ def test_rolling_median(self):
     def test_rolling_min(self):
         self._check_moment_func(mom.rolling_min, np.min)
 
+        a = np.array([1,2,3,4,5])
+        b = mom.rolling_min(a, window=100, min_periods=1)
+        assert_almost_equal(b, np.ones(len(a)))
+
+        self.assertRaises(ValueError, mom.rolling_min, np.array([1,2,3]), window=3, min_periods=5)
+
     def test_rolling_max(self):
         self._check_moment_func(mom.rolling_max, np.max)
 
+        a = np.array([1,2,3,4,5])
+        b = mom.rolling_max(a, window=100, min_periods=1)
+        assert_almost_equal(a, b)
+
+        self.assertRaises(ValueError, mom.rolling_max, np.array([1,2,3]), window=3, min_periods=5)
+
     def test_rolling_quantile(self):
         qs = [.1, .5, .9]