How to find most frequent values in numpy ndarray?

守給你的承諾、 提交于 2019-12-18 12:18:22

问题


I have a numpy ndarray with shape of (30,480,640), the 1th and 2th axis representing locations(latitude and longitute), the 0th axis contains actual data points.I want to use the most frequent value along the 0th axis at each location, which is to construct a new array with shape of (1,480,640).ie:

>>> data
array([[[ 0,  1,  2,  3,  4],
        [ 5,  6,  7,  8,  9],
        [10, 11, 12, 13, 14],
        [15, 16, 17, 18, 19]],

       [[ 0,  1,  2,  3,  4],
        [ 5,  6,  7,  8,  9],
        [10, 11, 12, 13, 14],
        [15, 16, 17, 18, 19]],

       [[40, 40, 42, 43, 44],
        [45, 46, 47, 48, 49],
        [50, 51, 52, 53, 54],
        [55, 56, 57, 58, 59]]])

(perform calculation)

>>> new_data 
array([[[ 0,  1,  2,  3,  4],
        [ 5,  6,  7,  8,  9],
        [10, 11, 12, 13, 14],
        [15, 16, 17, 18, 19]]])

The data points will contain negtive and positive floating numbers. How can I perform such calculations? Thanks a lot!

I tried with numpy.unique,but I got "TypeError: unique() got an unexpected keyword argument 'return_inverse'".I'm using numpy version 1.2.1 installed on Unix and it doesn't support return_inverse..I also tried mode,but it takes forever to process such large amount of data...so is there an alternative way to get the most frequent values? Thanks again.


回答1:


To find the most frequent value of a flat array, use unique, bincount and argmax:

arr = np.array([5, 4, -2, 1, -2, 0, 4, 4, -6, -1])
u, indices = np.unique(arr, return_inverse=True)
u[np.argmax(np.bincount(indices))]

To work with a multidimensional array, we don't need to worry about unique, but we do need to use apply_along_axis on bincount:

arr = np.array([[5, 4, -2, 1, -2, 0, 4, 4, -6, -1],
                [0, 1,  2, 2,  3, 4, 5, 6,  7,  8]])
axis = 1
u, indices = np.unique(arr, return_inverse=True)
u[np.argmax(np.apply_along_axis(np.bincount, axis, indices.reshape(arr.shape),
                                None, np.max(indices) + 1), axis=axis)]

With your data:

data = np.array([
   [[ 0,  1,  2,  3,  4],
    [ 5,  6,  7,  8,  9],
    [10, 11, 12, 13, 14],
    [15, 16, 17, 18, 19]],

   [[ 0,  1,  2,  3,  4],
    [ 5,  6,  7,  8,  9],
    [10, 11, 12, 13, 14],
    [15, 16, 17, 18, 19]],

   [[40, 40, 42, 43, 44],
    [45, 46, 47, 48, 49],
    [50, 51, 52, 53, 54],
    [55, 56, 57, 58, 59]]])
axis = 0
u, indices = np.unique(arr, return_inverse=True)
u[np.argmax(np.apply_along_axis(np.bincount, axis, indices.reshape(arr.shape),
                                None, np.max(indices) + 1), axis=axis)]
array([[ 0,  1,  2,  3,  4],
       [ 5,  6,  7,  8,  9],
       [10, 11, 12, 13, 14],
       [15, 16, 17, 18, 19]])

NumPy 1.2, really? You can approximate np.unique(return_inverse=True) reasonably efficiently using np.searchsorted (it's an additional O(n log n), so shouldn't change the performance significantly):

u = np.unique(arr)
indices = np.searchsorted(u, arr.flat)



回答2:


Use SciPy's mode function:

import numpy as np
from scipy.stats import mode

data = np.array([[[ 0,  1,  2,  3,  4],
                  [ 5,  6,  7,  8,  9],
                  [10, 11, 12, 13, 14],
                  [15, 16, 17, 18, 19]],

                 [[ 0,  1,  2,  3,  4],
                  [ 5,  6,  7,  8,  9],
                  [10, 11, 12, 13, 14],
                  [15, 16, 17, 18, 19]],

                 [[40, 40, 42, 43, 44],
                  [45, 46, 47, 48, 49],
                  [50, 51, 52, 53, 54],
                  [55, 56, 57, 58, 59]]])

print data

# find mode along the zero-th axis; the return value is a tuple of the
# modes and their counts.
print mode(data, axis=0)



回答3:


flatten your array, then build a collections.Counter from it. As usual, take special care when comparing floating-point numbers.



来源:https://stackoverflow.com/questions/12297016/how-to-find-most-frequent-values-in-numpy-ndarray

易学教程内所有资源均来自网络或用户发布的内容,如有违反法律规定的内容欢迎反馈
该文章没有解决你所遇到的问题?点击提问,说说你的问题,让更多的人一起探讨吧!