Suppose I have a tensor A (maybe it’s a free-variable covariance matrix). I would like to construct a block diagonal matrix with k repetitions of the matrix along the diagonal.
If I were working with numpy objects, I could achieve this with the code:
np.kron(np.eye(k), A)
Kroneker products are available in aesara, inside aesara.tensor.linalg:
import aesara.tensor as at
x = at.linalg.kron(at.eye(k), A)
There is also a helper function in PyMC for block-concatenation of a list of matrices, pm.math.block_diagonal, that functions similarly to scipy.linalg.block_diag
Thank you!
I do not see pm.math.block_diagonal in the API reference, though?
I guess it’s a documentation oversight! Here’s the function.