smallestProb = 1.0
docSet = set(data.docsPerQuery)
for i in docSet:
currentMin = None
if args.temperature > 0.0:
currentMin = numpy.amin(loggingPolicy.multinomials[i])
else:
currentMin = 1.0 / i
if currentMin < smallestProb:
smallestProb = currentMin
print("Parallel:main [LOG] Temperature:", args.temperature, "\t Smallest marginal probability:", smallestProb,
flush=True)
metric = None
if args.value_metric == "DCG":
metric = Metrics.DCG(data, args.length_ranking)
elif args.value_metric == "NDCG":
metric = Metrics.NDCG(data, args.length_ranking, args.replacement)
elif args.value_metric == "ERR":
metric = Metrics.ERR(data, args.length_ranking)
elif args.value_metric == "MaxRel":
metric = Metrics.MaxRelevance(data, args.length_ranking)
elif args.value_metric == "SumRel":
metric = Metrics.SumRelevance(data, args.length_ranking)
else:
print("Parallel:main [ERR] Metric %s not supported." % args.value_metric, flush=True)
sys.exit(0)
estimator = None
if args.approach == "OnPolicy":
estimator = Estimators.OnPolicy(args.length_ranking, loggingPolicy, targetPolicy, metric)
trainDataset.loadNpz(foldDir + 'train')
validationDataset = Datasets.Datasets()
validationDataset.loadNpz(foldDir + 'vali')
testDataset = Datasets.Datasets()
testDataset.loadNpz(foldDir + 'test')
loggingPolicy = Policy.UniformPolicy(trainDataset, False)
loggingPolicy.setupGamma(args.length_ranking)
trainMetric = None
validationMetric = None
testMetric = None
if args.value_metric == "DCG":
trainMetric = Metrics.DCG(trainDataset, args.length_ranking)
validationMetric = Metrics.DCG(validationDataset,
args.length_ranking)
testMetric = Metrics.DCG(testDataset, args.length_ranking)
elif args.value_metric == "NDCG":
trainMetric = Metrics.NDCG(trainDataset, args.length_ranking,
False)
validationMetric = Metrics.NDCG(validationDataset,
args.length_ranking, False)
testMetric = Metrics.NDCG(testDataset, args.length_ranking, False)
elif args.value_metric == "ERR":
trainMetric = Metrics.ERR(trainDataset, args.length_ranking)
validationMetric = Metrics.ERR(validationDataset,
args.length_ranking)
testMetric = Metrics.ERR(testDataset, args.length_ranking)
else:
