Please sign in to comment.
BUG: optimize: fix epsfcn choice in minpack routines
An additional sqrt() was introduced in 703dc23 --- minpack takes the sqrt internally, so it should not be done again. The routines can still fail if the problem is badly scaled. The test is here changed to be less badly scaled. The correct choice of epsfcn is problem-dependent and in general needs to be left to the user. Fixes gh-2639
- Loading branch information...
Showing with 23 additions and 4 deletions.