jax_privacy.matrix_factorization.dense

Optimization and error fns for dense (explicitly represented) strategies.

See sensitivity.py for sensitivity calculations for dense strategies.

Functions

get_orthogonal_mask(n[, epochs])	Computes a mask that imposes orthognality constraints on the optimization.
optimize(n, *[, epochs, bands, equal_norm, ...])	Optimizes a strategy matrix C for a given reduction_fn and participation.
per_query_error(*[, strategy_matrix, ...])	Expected per-query squared error for a general matrix mechanism.
pg_tol_termination_fn(step_info)	Callback function that returns True if projected gradient is near-zero.
strategy_from_X(X)	Return a lower triangular strategy matrix C from its Gram matrix.