TF 2.0 is out! Get hands-on practice at TF World, Oct 28-31. Use code TF20 for 20% off select passes. Register now

tf.linalg.logdet

TensorFlow 1 version 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.