how do I calculate a rolling idxmax

Question

consider the pd.Series s

import pandas as pd
import numpy as np

np.random.seed([3,1415])
s = pd.Series(np.random.randint(0, 10, 10         


        

Answer 1

Here's an approach using broadcasting -

maxidx = (s.values[np.arange(s.size-3+1)[:,None] + np.arange(3)]).argmax(1)
out = s.index[maxidx+np.arange(maxidx.size)]

This generates all the indices corresponding to the rolling windows, indexes into the extracted array version with those and thus gets the max indices for each window. For a more efficient indexing, we can use NumPy strides, like so -

arr = s.values
n = arr.strides[0]
maxidx = np.lib.stride_tricks.as_strided(arr, \
                   shape=(s.size-3+1,3), strides=(n,n)).argmax(1)