Closed
Description
I am updating a codebase from an OLD version of pandas (0.7.3) to current version (0.14.1) in python 2.7.
pandas: 0.14.1
numpy: 1.8.0
Not sure if this is properly classified as a bug, or if it is an expected code change. However, it seems rolling_cov / rolling_corr are specifying the window as min(window given, max(len(arg1), len(arg2)), and then erring out if arg1 AND arg2 are shorter than min_periods. Before, we would just get NaNs back.
Here is some test code:
import numpy as np
import pandas as pd
np.random.seed(0)
s1 = pd.Series(np.random.randn(4))
s2 = pd.Series(np.random.randn(4))
print '\nrolling_corr:'
try:
print pd.rolling_corr(s1, s2, window=10, min_periods=5)
except Exception as inst:
print type(inst)
print inst
print '\nrolling_cov:'
try:
print pd.rolling_cov(s1, s2, window=10, min_periods=5)
except Exception as inst:
print type(inst)
print instIn 0.14.1:
rolling_corr:
<type 'exceptions.ValueError'>
min_periods (5) must be <= window (4)
rolling_cov:
<type 'exceptions.ValueError'>
min_periods (5) must be <= window (4)
In 0.7.3:
rolling_corr:
0 NaN
1 NaN
2 NaN
3 NaN
rolling_cov:
0 NaN
1 NaN
2 NaN
3 NaN
Note that changing either s1 OR s2 to be 6, the code doesn't error out, and instead outputs:
rolling_corr:
0 NaN
1 NaN
2 NaN
3 NaN
4 NaN
5 NaN
dtype: float64
rolling_cov:
0 NaN
1 NaN
2 NaN
3 NaN
4 NaN
5 NaN
dtype: float64
Meanwhile, in 0.14.1, other rolling functions continue to behave as before:
s1 = pd.Series(np.random.randn(4))
print '\rolling_std:'
print pd.rolling_std(s1, window=10, min_periods=5)
print '\nrolling_max:'
print pd.rolling_max(s1, window=10, min_periods=5)Outputs:
rolling_std:
0 NaN
1 NaN
2 NaN
3 NaN
dtype: float64
rolling_max:
0 NaN
1 NaN
2 NaN
3 NaN
dtype: float64