Complete solves for QUBO

QUBO/default_config.csv

@@ -84,11 +84,11 @@ num_vertices,density,random_seed,c_optimal
 250,0.01,2,-137
 250,0.01,3,-138
 250,0.01,4,-143
-250,0.05,0,-67
-250,0.05,1,-69
+250,0.05,0,-69
+250,0.05,1,-70
 250,0.05,2,-67
-250,0.05,3,-67
-250,0.05,4,-67
+250,0.05,3,-70
+250,0.05,4,-68
 250,0.1,0,-43
 250,0.1,1,-44
 250,0.1,2,-44

QUBO/gurobi.py

@@ -0,0 +1,110 @@
+from qubo_mis_generator import MISBenchmark
+import csv
+import json
+import numpy as np
+
+from gurobipy import *
+
+
+def check_mis_np(Q, sample):
+    # check for independent set in order to ensure cost is minimized
+
+    assert len(sample) == qubo_matrix.shape[0], "Sample size does not match QUBO size"
+
+    # get the nodes that are part of the MIS
+    mis_nodes = [idx for idx in range(len(sample)) if sample[idx] == 1]
+
+    # check if the MIS is valid
+    for i in range(len(mis_nodes)):
+        for j in range(i + 1, len(mis_nodes)):
+            # check if dict entry exists
+            if qubo_matrix[mis_nodes[i], mis_nodes[j]] != 0:
+                return False
+
+    return True
+
+def check_mis_bqm(bqm, sample):
+    # check for independent set in order to ensure cost is minimized
+
+    assert len(sample) == bqm.num_variables, "Sample size does not match BQM size"
+
+    # get the nodes that are part of the MIS
+    mis_nodes = [node for node in sample if sample[node] == 1]
+
+    # check if the MIS is valid
+    for i in range(len(mis_nodes)):
+        for j in range(i + 1, len(mis_nodes)):
+            # check if dict entry exists
+            if bqm.adj.get(mis_nodes[i]).get(mis_nodes[j]) is not None:
+                return False
+
+    return True
+
+def sample_to_binary_list(sample):
+    N = len(sample)
+    mis_nodes = [node for node in sample if sample[node] == 1]
+    l = [0 if i not in mis_nodes else 1 for i in range(N)]
+    return l
+
+if __name__ == '__main__':
+    benchmark = MISBenchmark('default_config.csv')
+    dataset_dir = './data/qubo_mis_dataset'
+
+    # look for temp costs file
+    try:
+        with open('costs.json', 'r') as f:
+            costs = json.load(f)
+    except FileNotFoundError:
+        costs = {}
+
+    for qubo_matrix, num_vertices, density, seed, _ in benchmark.load_problems(dataset_dir):
+        print("Problem: ", num_vertices, density, seed)
+        key = f'{num_vertices}_{density}_{seed}'
+
+        if key in costs:
+            continue
+
+        if num_vertices < 1000:
+            num_nodes = qubo_matrix.shape[0]
+            quadratic_model = Model()
+
+            x = quadratic_model.addMVar(num_nodes, vtype=GRB.BINARY, name='x')
+            quadratic_model.setObjective(x @ qubo_matrix @ x)
+
+            quadratic_model.optimize()
+
+            c_optimal = quadratic_model.objVal
+            sample = x.X
+
+            # sanity check
+            assert check_mis_np(qubo_matrix, sample), "Gurobi did not find a valid MIS"
+
+            costs[key] = int(c_optimal)
+
+        # else:
+        #     bqm = BinaryQuadraticModel(qubo_matrix, "BINARY")
+
+        #     sampler = TabuSampler()
+        #     sampleset = sampler.sample(bqm, timeout=None, num_reads=100, num_restarts=50)
+
+        #     # solver sanity check
+        #     # while not guaranteed, it is highly likely the solver finds independent sets
+        #     assert check_mis_bqm(bqm, sampleset.first.sample), "Solver did not find a valid MIS"
+
+        #     print("found min cost", int(sampleset.first.energy), "for sample", sample_to_binary_list(sampleset.first.sample), flush=True)
+
+        #     costs[key] = int(sampleset.first.energy)
+
+        # temporary save costs dict as json
+        with open('costs.json', 'w') as f:
+            json.dump(costs, f)
+
+    # rewrite csv
+    header = ['num_vertices', 'density', 'random_seed', 'c_optimal']
+    with open('gurobi_default_config.csv', 'w', newline='') as f:
+        writer = csv.writer(f)
+        writer.writerow(header)
+        for key, value in costs.items():
+            num_vertices, density, random_seed = key.split('_')
+            c_optimal = value
+            writer.writerow([num_vertices, density, random_seed, c_optimal])