How to find maximum of previous x elements using numpy

user_5426513 0
Viewed 80

Question

Is there a way using numpy to sort maximum of range of previous x numbers?


Example

Array of [1, 3, 4, 2, 6, 7, 8, 8, 2, 3], and the range is 3

The output should be [nan, nan, nan, 4, 4, 6, 7, 8, 8, 8].

The first three elements has no 3 previous numbers, so the result is nan.

The fourth elements (2) has the previous numbers of 1, 3, 4, therefore the max is 4.

Sorry for the misleading.

I will provide a second example.

Original array: [1, 3, 4, 6, 2, 8, 2, 3, 1, 2] Output: [nan, nan, nan, 4, 6, 6, 8, 8, 8, 3]

For the last element of output: 3, the reason in the 7th, 8th and 9th elements in original array are 2, 3, 1, therefore the max is 3.

Thanks!

pandas dataframe numpy python
3 Answers

Here is a pandas solution:

win_size = 3
lst = [1, 3, 4, 2, 6, 7, 8, 8, 2, 3]

result = pd.Series(lst).rolling(win_size).max().shift(1).tolist()

Output:

[nan, nan, nan, 4.0, 4.0, 6.0, 7.0, 8.0, 8.0, 8.0]

Note: Outputs are floats - if that bothers you, feel free to convert.

user_129833990

You can't have nan in an np.array that is of data type int, so you can do the following. Change dtype to .astype(float). Set the first 3 values to the minimum of the array minus 1. Sort the values. Change the new minimum values of the first 3 numbers to np.nan.

Note, sorting with np.nan, puts the nan values at the end, which is why I didn't change the values until after sorting.

a = np.array([1, 3, 4, 2, 6, 7, 8, 8, 2, 3]).astype(float)
a[np.arange(3)] = np.min(a-1)
a=np.sort(a)
a[np.arange(3)] = np.nan
a

array([nan, nan, nan,  2.,  2.,  3.,  6.,  7.,  8.,  8.])
user_63667700

You can use maximum.accumulate to get an iterative max

import numpy as np

a = np.array([1, 3, 4, 2, 6, 7, 8, 8, 2, 3])
a = np.maximum.accumulate(a).astype('float')

Converting to float is necessary to set values to np.nan in a numpy array. This clears the unwanted values.

a[np.arange(3)] = np.nan
a

Out:

array([nan, nan, nan,  4.,  6.,  7.,  8.,  8.,  8.,  8.])

For shifted rolling max with numpy if pandas is not an option

a =  np.array([1, 3, 4, 2, 6, 7, 8, 8, 2, 3])
window = 3
rolling = np.lib.stride_tricks.as_strided(a, shape=(len(a)-window, window), strides=a.strides + a.strides)
np.r_[[np.nan]*window, np.max(rolling, axis=1)]

Out:

array([nan, nan, nan,  4.,  4.,  6.,  7.,  8.,  8.,  8.])
user_142777220
Related