def Kbundle_QP_relaxation(mu_vec, Cov_mat, alpha, K, separate_profit_vec,
mandatory_products):
N = len(mu_vec)
# M is the parameter for the number of candidates. When M is small, the remaining gap is more important and reducing the variance is more important when M is large.
gap_vec = [None] * N
for i in range(N):
gap_vec[i] = (mu_vec[i] - separate_profit_vec[i])
sorted_indices = [
i[0]
for i in sorted(enumerate(gap_vec), key=lambda x: x[1], reverse=True)
]
# Insere os produtos obrigatórios no início do índice de produtos mais lucrativos
sorted_indices = [x for x in sorted_indices if x not in mandatory_products]
for product_index in mandatory_products:
sorted_indices.insert(0, product_index)
max_profit = -math.inf
max_bundle = None
chunk = 20
opt_profit_vec = []
for M in range(K, min(3 * K + 1, N), max(2 * K // chunk, 1)):
# print ('====considering the first', M, 'product with highest gap')
opt_soln = Kbundle_candidate(sorted_indices[:M], Cov_mat, K,
len(mandatory_products))
# inner sampling
opt_profit = -math.inf
opt_bundle = None
sample_size = 5 # we do randomized rounding for 5 times and obtain the best
for i in range(sample_size):
bundle_set, bundle_vec = random_set(opt_soln, K)
variance_bundle = profit.variance(bundle_vec, Cov_mat)
variance_opt = profit.variance(opt_soln, Cov_mat)
# print (pi.bcolors.FAIL,'opt=', variance_opt, 'bundle=', variance_bundle , pi.bcolors.ENDC)
profit_ = profit.profit_bundle(bundle_set, bundle_vec, mu_vec,
Cov_mat, alpha, separate_profit_vec)
if profit_ > opt_profit:
opt_profit = profit_
opt_bundle = bundle_set
opt_profit_vec.append(opt_profit)
if opt_profit > max_profit:
max_profit = opt_profit
max_bundle = opt_bundle
return max_bundle, max_profit
def opt_enumerate(mean_vec, cov_mat, k, separate_profit_vec):
alpha = 0 # dummy
N = len(mean_vec)
products = list(range(N))
max_profit = -math.inf
opt_bundle = None
for bundle_set in itertools.combinations(products, k):
bundle_vec = [0] * N
for i in bundle_set:
bundle_vec[i] = 1
revenue = profit.profit_bundle(bundle_set, bundle_vec, mean_vec, cov_mat, alpha, separate_profit_vec)
if revenue > max_profit:
max_profit = revenue
opt_bundle = bundle_set
return float(max_profit)
bundle_size_vec = exp_setting.bundle_size_vec
profit_vec = []
sample_size = 50
for k in bundle_size_vec:
print('bundle size:', k)
sample_revenue_vec = []
for s in range(sample_size):
choices = np.random.choice(N, k)
bundle_set = set(choices)
bundle_vec = [0] * N
for i in range(N):
if i in bundle_set:
bundle_vec[i] = 1
revenue = profit.profit_bundle(bundle_set, bundle_vec, mean_vec,
cov_mat, alpha, separate_profit_vec)
print('profit:', revenue)
sample_revenue_vec.append(revenue)
profit_vec.append(sum(sample_revenue_vec) / len(sample_revenue_vec))
with open('data/result_baseline_random_profits.json', 'w') as output_file:
json.dump(
{
"bundle_size_vec": bundle_size_vec,
"profit_vec": profit_vec
},
output_file,
indent=4)