Open gwding opened 5 years ago
Hi,
This output looks fine to me, and should be the same as (or at least similar to) what using Gurobi will give you. Gurobi may be a bit faster in certain cases though, which could decrease the number of samples the verifier times out on.
@kaixiao That's great! thx!
Having another question: for Gurobi, there's a parameter BestObjStop=eps
. I cannot find the same parameter in Cbc. Is this just an "early stopping" criterion?
Thanks for making the code available!
I wonder if we can use opensource solvers other than gurobi? I tried to modify
verify_MNIST.jl
to work with the Cbc solver. I changed a few lines as belowIt seems working fine, and the output
summary.csv
file isCould you help verify if this is working as expected? Thanks!