projnormal.param_sampling

Functions for randomly sampling distribution parameters.

Functions

make_mean(n_dim[, shape, sparsity])

Generate a vector to use as the mean of a multivariate normal.

make_ortho_vectors(n_dim, n_vec)

Generate a set of orthogonal vectors.

make_spdm(n_dim[, eigvals, eigvecs])

Make a symmetric positive definite matrix.
make_mean(n_dim, shape='gaussian', sparsity=0.1)

Generate a vector to use as the mean of a multivariate normal.

Parameters:

  • n_dim (int) – Dimension of the mean vector.

  • shape (str) – Type of mean vector. Options are: 'gaussian' (each element sampled from N(0,1)), 'exponential' (each element sampled from Exp(1)), 'sin' (sin-wave vector with random phase frequency and amplitude), 'sparse' (sparse vector with 0s and 1s)

  • sparsity (float) – For 'sparse' shape, the fraction of non-zero elements

Returns:

Mean vector. Shape is (n_dim,).

Return type:

torch.Tensor

make_ortho_vectors(n_dim, n_vec)

Generate a set of orthogonal vectors.

Parameters:

  • n_dim (int) – Dimension of the vectors.

  • n_vec (int) – Number of orthogonal vectors to generate. Must be less than n_dim.

Returns:

Orthogonal vectors of size n_dim x n_vec. Shape is (n_vec, n_dim).

Return type:

torch.Tensor

make_spdm(n_dim, eigvals='uniform', eigvecs='random')

Make a symmetric positive definite matrix.

Parameters:

  • n_dim (int) – Dimension of matrix

  • eigvals (str or torch.Tensor) – Eigenvalues of the matrix. Options are: 1) Tensor of eigvals to use, of length n_dim. 2) ‘uniform’: Eigvals are uniformly sampled between 0.1 and 1. 3) 'exponential': Eigvals sampled from Exp(1).

  • eigvecs (str) – Eigenvectors of the matrix. Options are: 1)``’random’: Random orthogonal matrix. 2) ``'identity': Identity matrix.

Returns:

Symmetric positive definite matrix with specified eigvals. Shape is (n_dim, n_dim).

Return type:

torch.Tensor