tf_agents.bandits.policies.linalg.conjugate_gradient_solve

View source on GitHub

Returns X such that A * X = B.

tf_agents.bandits.policies.linalg.conjugate_gradient_solve(
    a_mat, b_mat, tol=1e-10
)

Uses Conjugate Gradient to solve many linear systems of equations with the same matrix a_mat and multiple right hand sides provided as columns in the matrix b_mat.

Args:

• a_mat: a Symmetric Positive Definite matrix, represented as a Tensor of shape [n, n].
• b_mat: a Tensor of shape [n, k].
• tol: (float) desired tolerance on the residual.

Returns:

• X: Tensor X of shape [n, k] such that A * X = B.

Raises:

• ValueError: if a_mat is not square or a_mat and b_mat have incompatible shapes.