How to find most frequent values in numpy ndarray?

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)

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)

A slightly better solution in my opinion is the following

tmpL = np.array([3, 2, 3, 2, 5, 2, 2, 3, 3, 2, 2, 2, 3, 3, 2, 2, 3, 2, 3, 2])
unique, counts = np.unique(tmpL, return_counts=True)
return unique[np.argmax(counts)]

Using np.unique we can get the count of each unique elements. The index of the max element in counts will be the corresponding element in unique.