Numpy 按行掩码

Question

I have 2D array:

matrix =np.array([[95,90,-1,55],[100,90,-1,80],[0,90,85,100]])

I try to choose 2 random rows, and ignore -1.

I tried:

random_ints = np.random.choice(len(matrix), size=2, replace=False)
students = matrix[random_ints, :]
ignored = students[students != -1]

but the ignored var reshaped to 1D array for example:

[95,90,55,100,90,80]

I want the ignored saved 2D for example:

[[95,90,55],[100,90,80]]

How can do that?

Answer 1

If you had equal number of elements to be left ( != -1 ) in each row, you could run:

np.apply_along_axis(lambda row: row[row != -1], 1, students)

To check this variant, set random_ints = np.array([0,1]) , select just the above rows to students and run the above code.

But if the number of these element is different , Numpy will raise exception:

ValueError: could not broadcast input array from shape (3) into shape (4)

The reason is that (in this case) some rows have 4 elements to be left, whereas other - only 3 .

Unfortunately, Numpy does not support "jagged" arrays (with different number of elements in each row).

You can get something like you wanted, but as a nested list (not a Numpy array):

result = []
for row in students:
    result.append(row[row != -1].tolist())

The result (for rows 2 and 1 ) is:

[[0, 90, 85, 100], [100, 90, 80]]