In practice, when would you want to use finite differences instead of some flavour of AD? I've found FD to be very brittle, and only really useful to check that my analytical derivatives are correct
-
-
-
It's a nice default if you don't know if something is AD-compatible. FWIW, this same sparsity support exists in SpraseDiffTools.jl in a way that uses ForwardDiff.jl for sparse/structured forward-mode AD Jacobians.
- Još 2 druga odgovora
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.