np.expand_dims takes tuples

[chillenzer]

Replace the two calls to np.expand_dims with one call with axis=<tuple>, i.e.

matrix_weights = np.expand_dims(
    [[0, 2, 0], [1, 0, 3]], axis=(2, 3)
)
values_4d * matrix_weights

instead of the currently present

matrix_weights = np.expand_dims(np.expand_dims(
    [[0, 2, 0], [1, 0, 3]], axis=2), axis=3
)
values_4d * matrix_weights

in Broadcasting.

[chillenzer]

There are more instances of this in the episode. Check carefully!