bug - for a large number of points

[haoqian-hao]

I tried running this with 20 points on D'wave, but it gave me an error:
ValueError: no embedding found
is this because dwavebinarycsp can not solve a large dataset?

[arcondello]

Hi @haoqian-hao , probably the BQM that is made by dwavebinarycsp is too large for the QPU you are using. You should consider using one of our hybrid solvers.

[haoqian-hao]

Hi @arcondello, when I was using my dataset (the fraction that is laying on 0-1), it is not working correctly, it will assign the data point to all clusters (one of the results shows "0.15625,0_b': 1, '0.15625,0_g': 1, '0.15625,0_r': 1" ), it seems like the constraints in csp that penalise the coordinate can only be in one colour group is not strong enough, are there some ways to solve this? e.g., add chain_strength to the hybrid solver or add a coefficient or a penalty strength?

I appreciate any help you can provide.

[arcondello]

Which hybrid solver are you using? And are you still using dwavebinarycsp.stitch() to construct your BQM?

[haoqian-hao]

@arcondello, I used LeapHybridSampler and got the lowest energy from the result followed by,

sampler = LeapHybridSampler()    
sampleset = sampler.sample(bqm) 
best_answer = list(sampleset.data(["sample", "energy"]))[0].sample

Yes, I am using dwavebinarycsp.stitch() to construct BQM

[arcondello]

Hi @haoqian-hao, you can adjust the classical gap for the penalty models with dwavebinarycsp.stitch(csp, min_classical_gap=...), see min_classical_gap.

You might also try using our constrained quadratic model hybrid solver.

[haoqian-hao]

Hi @arcondello, thank you for helping me for this long time.
I just wondering why

BQM that is made by dwavebinarycsp is too large for the QPU you are using

from my understanding, the qubits require for solving Clustering are
K * N, where K = the number of clusters and N is datapoints,
for the pure QPU like DW_2000Q_6, it has 2041 working qubits. Therefore, the DW_2000Q_6 should be able to handle ( N <= 2041 / K ) datapoint. However, from my testing, it can only handle ~10 for K=2. just wondering why this happened, the reason could be we can not fully access all of the qubits? just wondering should we set the hyperparameter which represents the number of qubits used?

[arcondello]

Hi @haoqian-hao, although the QPU has ~2000 working qubits, they are not all connected to eachother. Therefore, solving problems with higher connectivity requires something called embedding. See D-Wave QPU Architecture: Topologies for a discussion of what this means for solving problems.

Also worth noting that the hybrid solvers do not have these limits.

[billwisotsky]

Hello. I have been playing around with this sample. Once I go above 18 tuples I get the following error. Not sure what this means. The sampler, sampleset and best_sampler have not been changed from what is in the example,

Thanks in advance.

ValueError Traceback (most recent call last)
Input In [18], in <cell line: 2>()
1 clustered_filename = "clustered_points.png"
----> 2 cluster_points(scattered_points, clustered_filename, False)

File ~\OneDrive - SAS\Quantum\DWave\Notebooks\clustering-master\clustering_SAS.py:89, in cluster_points(scattered_points, filename, problem_inspector)
86 weight = -math.cos(d*math.pi)
88 # Apply weights to BQM
---> 89 bqm.add_interaction(coord0.r, coord1.r, weight)
90 bqm.add_interaction(coord0.g, coord1.g, weight)
91 bqm.add_interaction(coord0.b, coord1.b, weight)

File ~.conda\envs\DWave\lib\site-packages\dimod\binary\binary_quadratic_model.py:888, in BinaryQuadraticModel.add_interaction(self, *args, **kwargs)
886 def add_interaction(self, *args, **kwargs):
887 """Alias for :meth:.add_quadratic."""
--> 888 return self.add_quadratic(*args, **kwargs)

File ~.conda\envs\DWave\lib\site-packages\dimod\binary\cybqm\cybqm_template.pyx.pxi:249, in dimod.binary.cybqm.cybqm_float64.cyBQM_template.add_quadratic()

ValueError: '51.0,38.0_r' cannot have an interaction with itself

[billwisotsky]

All fixed. Worked with Dwave today and looks like since I using integers, there was another pair with the same values which through the error. Once I added decimal places I got past this error.

[billwisotsky]

Has anyone gotten this to work with large samples. I have run through this an tried the potential fixes mentioned above (i.e. min_classical_gap) and nothing has worked. min_classical_gap would just generate errors below. It seems anything over sample size of 9 per cluster and the clustering breaks down. Attached is a graph with 100. There are visually 3 clusters, but they are only being assigned to one cluster name red.

Traceback (most recent call last):
File "c:/Users/bill/OneDrive/Quantum/DWave/Notebooks/clustering-master/example_clusters.py", line 53, in

cluster_points(scattered_points, clustered_filename, args.problem_inspector)

File "c:\Users\bill\OneDrive\Quantum\DWave\Notebooks\clustering-master\clustering.py", line 124, in cluster_points
groupings = get_groupings(best_sample)
File "c:\Users\bill\OneDrive\Quantum\DWave\Notebooks\clustering-master\utilities.py", line 36, in get_groupings
coord, color = label.split("_")
ValueError: not enough values to unpack (expected 2, got 1)