Using Leap to solve a 100-vertex Max-Cut problem but not getting accurate results?
Hello,
I am attempting to solve a 100-vertex max cut graph by adjusting the max-cut code given here (https://github.com/dwave-examples/maximum-cut) and inserting my own pre-built matrix, but even after increasing numruns to 3000 and chain_strength to 8, I am only getting a response that is 89.4% accurate after about ~25 seconds of run time.
Meanwhile, using Dwave's simulated annealing program (https://docs.ocean.dwavesys.com/projects/neal/en/latest/reference/sampler.html) to solve the same matrix, I get a response that is 100% accurate after just 0.01 seconds.
Is this the optimal quantum response we could expect from using Leap Advantage for a 100-vertex graph (i.e. <100% accuracy), or does my quantum annealing code need further tweaking? Are there other solvers or methods I should consider using?
For reference, I am using the sample_ising() solver to solve the quantum problem.
Thanks and Best Regards,
Rob
Comments
Hi Robert,
Increasing the number of reads was a good idea, that's usually where I start.
I'm curious about the chain_strength value, when you print the results, what does it show for the chain_break_fraction (chain_.) column? Does it show values > 0?
What's the magnitude in the pre-built matrix? Normally, the starting point for chaing_strength is the magnitude of the linear or quadratic terms. From there, we may be able to tweak results by increasing chain_strength.
Please sign in to leave a comment.