Timeout error

Subarno S

July 16, 2020 00:51

I am attempting to use a hybrid solver to solve an optimization problem with over 10,000 variables. I am doing this on my local machine. I am aware that the generic leap hybrid solver does not allow for this so I tried using a customizable workflow hybrid solver with racing tabu sampler and qpu sampler. I was also able to break this problem down into chunks so I believed that this custom hybrid solver could handle it. Instead It took excessively long to run (i suspect that this was because of the length of time which was taken to translate my Hamiltonian) and finally it ended with a connection or timeout error. I took care to ensure that my computer did not sleep on execution but this still happened. Each time after the error would occur I would make sure that I had not hit my usage limits. As of this writing despite this happening on multiple occasions i have only used 13% of my allocated qpu time. While the translation takes time I am not inclined to believe that the translation to a Hamiltonian is a problem as each time i execute, the amount of allocated QPU time decreases suggesting that it does successfully send the Hamiltonian for computation to the hybrid solver. In addition I don't believe that the problem that I have given my hybrid sampler is too difficult as I have executed this optimization problem multiple times before using classic optimization. In those instances it has only taken a few minutes at most for the optimization to complete.

Comments

3 comments

Alex K (Report)
July 16, 2020 18:11
It sounds like there are a couple issues going on here
1. The long run time
2. The timeout or connection error
Regarding the long run time, dwave-hybrid has not been optimized to handle large problems like yours yet. This could explain the slow processing time, however it should not cause issues with your connection or the completion of your problem. Can you share more details about your hybrid workflow? How are you translating your problem to a Hamiltonian and how are you decomposing it? It would also be helpful if you can post your code and the version of Ocean you're using.

For the connection issue, how stable is your internet? Even if your computer doesn't go to sleep if you have any network issues they could cause an error like you're describing. Can you post the error message?

It may also be worth checking your Leap Dashboard to see when the latest hybrid problem submission was. There will be a separate problem ID for each subproblem your hybrid framework submits to the QPU. That may give you a clue as to when the connection timed out.
0

Comment actions Permalink
Subarno S (Report)
July 21, 2020 07:30 (Edited July 21, 2020 07:35)
I can confirm that my computer is not going to sleep. I can also confirm that my computer never lost internet connection between when I first executed it and when it gave me that error.

as for my code:
```
workflow = hybrid.Loop(
    hybrid.Race(
        hybrid.InterruptableTabuSampler(),
        hybrid.EnergyImpactDecomposer(size=10, rolling=True, traversal='bfs')
        | hybrid.QPUSubproblemAutoEmbeddingSampler()
        | hybrid.SplatComposer()) | hybrid.ArgMin(), convergence=3)
```
a96bf848-a60d-4c76-93ca-eeb3b82e2d98

Thu, Jul 9, 2020 7:42 AM

07:42:54 AM

Completed

cebe77a1-e0f4-46a5-af19-56ec197828da

Thu, Jul 9, 2020 7:33 AM

07:33:10 AM

Completed

2a20f229-1956-453c-96ed-daac41dfce9a

Thu, Jul 9, 2020 7:23 AM

07:23:35 AM

Completed

880adddf-2392-4688-9804-af78e09c02be

Thu, Jul 9, 2020 7:17 AM

07:17:54 AM

Completed

02546a6b-0dfd-4545-bb04-d763838716dd

Thu, Jul 9, 2020 7:12 AM

07:12:24 AM

Completed
0

Comment actions Permalink

Subarno S (Report)

July 30, 2020 16:15

sorry for the delay. Here is my code:

for u in g.nodes:#range(20639):#
#if np.where(lossa != r):#20639
if u == 0:
#print("a")
u=1
if u != 20640:
#Q.append(lossa[u])
v = u-1
#print(lossa[0,u])
#print("u")
#print(u)
#print(lossa[u])
#print("v")
#print(v)
#print(lossa[v])

Q[(0,u)] += lossa[0,u]
#print(u) 
Q[(v,u)] += 10000
else:
break
print(Q)

workflow = hybrid.Loop(
hybrid.Race(
hybrid.InterruptableTabuSampler(),
hybrid.EnergyImpactDecomposer(size=1000, rolling=True, traversal='bfs')
| hybrid.QPUSubproblemAutoEmbeddingSampler()
| hybrid.SplatComposer()) | hybrid.ArgMin(), convergence=3)

result = hybrid.HybridSampler(workflow).sample_qubo(Q,chain_strength=chain_strength,num_reads=5)

Please sign in to leave a comment.

Didn't find what you were looking for?

New post