Revision history [back]

The ValueError regarding incompatible shapes between (None, 20, 9) and (None, 9) can be resolved by reshaping the (None, 9) array to (None, 1, 9) so that it can be broadcasted with the (None, 20, 9) array in the operation.

You can use the np.newaxis attribute to achieve this. Here's an example:

import numpy as np

# sample arrays
a = np.random.rand(10, 20, 9)
b = np.random.rand(10, 9)

# reshape b to (None, 1, 9)
b = np.reshape(b, (b.shape[0], 1, b.shape[1]))

# perform operation
c = a * b

In the above code, we first reshape the b array using np.reshape to add the new axis of size 1 in the second dimension. This results in the shape of (None, 1, 9).

We then perform the operation between a and b, resulting in the shape of (None, 20, 9) which is broadcasted over the (None, 1, 9).

This should resolve the ValueError issue.