tf.linalg.logdet

View source on GitHub

Computes log of the determinant of a hermitian positive definite matrix.

Aliases:

• tf.compat.v1.linalg.logdet
• tf.compat.v2.linalg.logdet

tf.linalg.logdet(
    matrix,
    name=None
)

# Compute the determinant of a matrix while reducing the chance of over- or
underflow:
A = ... # shape 10 x 10
det = tf.exp(tf.linalg.logdet(A))  # scalar

Args:

• matrix: A Tensor. Must be float16, float32, float64, complex64, or complex128 with shape [..., M, M].
• name: A name to give this Op. Defaults to logdet.

Returns:

The natural log of the determinant of matrix.

Numpy Compatibility

Equivalent to numpy.linalg.slogdet, although no sign is returned since only hermitian positive definite matrices are supported.