ksddescent.ksdd_lbfgs¶
- ksddescent.ksdd_lbfgs(x0, score, kernel='gaussian', bw=1.0, max_iter=10000, tol=1e-12, beta=0.5, store=False, verbose=False)¶
Kernel Stein Discrepancy descent with L-BFGS
Perform Kernel Stein Discrepancy descent with L-BFGS. L-BFGS is a fast and robust algorithm, that has no critical hyper-parameter.
- Parameters
- x0torch.tensor, size n_samples x n_features
initial positions
- scorecallable
function that computes the score
- kernl‘gaussian’ or ‘imq’
which kernel to choose
- max_iterint
max numer of iters
- bwfloat
bandwidth of the stein kernel
- tolfloat
stopping criterion for L-BFGS
- storebool
whether to stores the iterates
- verbose: bool
wether to print the current loss
- Returns
- x: torch.tensor, size n_samples x n_features
The final positions
References
A.Korba, P-C. Aubin-Frankowski, S.Majewski, P.Ablin. Kernel Stein Discrepancy Descent International Conference on Machine Learning, 2021.