search

lululxvi / deepxde

A library for scientific machine learning and physics-informed learning
https://deepxde.readthedocs.io
GNU Lesser General Public License v2.1
2.76k stars 758 forks source link

Backend paddle: switch default line_search_fn of L-BFGS to 'strong_wolfe' #1790

Closed lijialin03 closed 4 months ago

lijialin03 commented 4 months ago
  1. switch default line_search_fn of L-BFGS to 'strong_wolfe'
  2. 'n_iter' in the L-BFGS of paddle and torch will be increased by one at the beginning. Therefore, the original code cannot achieve the expected effect of exiting after training converges because 'prev_n_iter' always equal to 'n_iter - 1' in that line of code. Fixed it in the PR (by the way, for backend torch).