熊猫rolling_apply用于列表元素

Question

I have accumulated counter, whose elements are lists. Each list element is accumulated, compared to the previously row.

import pandas as pd
d=pd.DataFrame({'counter': {0: [1,0,2], 1:[1,2,3], 2:[4, 4, 5]}})

I can get the maxcounter using the apply function.

def maxf(x): return max(x.counter)
d['maxcounter']=d.apply(lambda row: maxf(row), axis=1)  

Now I would also like to have the field "max_increment", get the difference between current row and the previously row, then compute the maximum. Could we use rolling_apply for this?

The expected output is as below.

     counter    maxcounter  increase_max
0   [1, 0, 2]   2   NaN 
1   [1, 2, 3]   3   2
2   [4, 4, 5]   5   3

Note: counter is a list, each element is incremented from separated sensor. This is not optimized structure, but what we get now.

     counter    counter_incr    increase_max   max_incr_index
0   [1, 0, 2]   
1   [1, 2, 3]   [0, 2, 1]         2                1
2   [4, 4, 5]   [3, 2, 2]         3                0

Answer 1

get the difference between current row and the previously row

Since your type is not numeric (it is a list), it seems the easiest thing is to perform a pd.Series.shift followed by an operation subtracting each element from the previous one:

import numpy as np
>>> [(np.array(c) - np.array(p)) \
    for c, p in zip(d.counter, d.counter.shift(-1))]
[array([ 0, -2, -1]), array([-3, -2, -2]), array([ nan,  nan,  nan])]