Open disacrol opened 2 years ago
This value is machine-specific, curious why it would be too small to work? It just represents the h
value in the standard derivative formula needed to calculate the Jacobian:
Problem is in a scenario where the objective function is not a closed-form expression - instead it corresponds to the concatenation of several independent blocks, and the final results do not change with such small epsilon value in the Jacobian formula you mention there.
Hi @disacrol this should be doable. I'd welcome a PR from you that allowed setting epsilon!
@jamesasefa I am onto it. But I'll take the chance to revert again the changes done by @pkernevez in PR #12. Otherwise I won't be able to proceed as I work on windows. For more details check Issue #15.
Thanks,
https://github.com/skew-opensource/slsqp4j/blob/443dfb3e45e684385fdde30a5790db80cfa64cbb/slsqp4j/src/main/java/com/skew/slsqp4j/Jacobian.java#L51
Could this be editable? Have a use case for this algorithm where this default value is too small to work. Or is there a different, more adequate way to effectively use a different "epsilon" value?