# tf.contrib.kfac.fisher_factors.compute_cov

tf.contrib.kfac.fisher_factors.compute_cov(
tensor,
tensor_right=None,
normalizer=None
)


Compute the empirical second moment of the rows of a 2D Tensor.

This function is meant to be applied to random matrices for which the true row mean is zero, so that the true second moment equals the true covariance.

#### Args:

• tensor: A 2D Tensor.
• tensor_right: An optional 2D Tensor. If provided, this function computes the matrix product tensor^T * tensor_right instead of tensor^T * tensor.
• normalizer: optional scalar for the estimator (by default, the normalizer is the number of rows of tensor).

#### Returns:

A square 2D Tensor with as many rows/cols as the number of input columns.