TypeError: To be compatible with tf.eager.defun, Python functions must return zero or more Tensors; ... *As this function is running in eager mode prefectly fine, i am wondering is it a strict ...

Using the scipy.linalg.lu function as an example, if I run this in python with a 3x3 array as input, I get a tuple of 3 3x3 arrays returned, or if I wrap the return value with np.array() I get a 3x3x3 ...