ranking multiple numpy arrays

You could use np.argpartition to find the smallest 2 values for each column:

import numpy as np
p1  = np.array([140,142,145])
p2  = np.array([130,144,147])
p3  = np.array([150,141,147])
p4  = np.array([150,141,148])
P = np.row_stack([p1,p2,p3,p4])

result = np.argpartition(P, 2, axis=0) < 2
print(result)

yields

[[ True False  True]
 [ True False  True]
 [False  True False]
 [False  True False]]

np.argpartition(arr, k) partially sorts arr in ascending order. Each group of k elements is smaller than the next group of k elements, but within each group the elements may not be sorted.

Note that the code above always has exactly 2 True values per column. It finds 2 of the lowest values for each column, but may not find all such values. If you wish to find all such values, you could use

In [302]: P <= P[np.argpartition(P, 2, axis=0), np.arange(P.shape[1])][1]
Out[302]: 
array([[ True, False,  True],
       [ True, False,  True],
       [False,  True,  True],
       [False,  True, False]], dtype=bool)

P[np.argpartition(P, 2, axis=0), np.arange(P.shape[1])] returns P in column-sorted order.

In [5]: P[np.argpartition(P, 2, axis=0), np.arange(P.shape[1])]
Out[5]: 
array([[130, 141, 145],
       [140, 141, 147],
       [150, 142, 147],
       [150, 144, 148]])

P[np.argpartition(P, 2, axis=0), np.arange(P.shape[1])][1] selects the 2nd row. These are the 2nd lowest values in each column.

In [6]: P[np.argpartition(P, 2, axis=0), np.arange(P.shape[1])][1]
Out[6]: array([140, 141, 147])

The comparison P <= np.array([140, 141, 147]) is performed by broadcasting the array on the right-hand side from shape (3,) up to shape (4,3) so the comparison can be done element-wise.