ott.solvers.nn.conjugate_solvers.FenchelConjugateLBFGS#

class ott.solvers.nn.conjugate_solvers.FenchelConjugateLBFGS(gtol=0.001, max_iter=10, max_linesearch_iter=10, linesearch_type='backtracking', decrease_factor=0.66, ls_method='strong-wolfe')[source]#

Solve for the conjugate using jaxopt.LBFGS.

Parameters:

  • gtol (float) – gradient tolerance

  • max_iter (int) – maximum number of iterations

  • max_linesearch_iter (int) – maximum number of line search iterations

  • linesearch_type (Literal['zoom', 'backtracking']) – type of line search

  • decrease_factor (float) – decrease factor for a backtracking line search

  • ls_method (Literal['wolf', 'strong-wolfe']) – the line search method

Methods

solve(f, y[, x_init])

Solve for the conjugate.

Attributes

decrease_factor

gtol

linesearch_type

ls_method

max_iter

max_linesearch_iter