def read_bandit_data(self, repo_dir, instance_type):
if repo_dir.endswith('.txt'):
fileName = repo_dir
else:
fileName = repo_dir + instance_type + '.txt'
if not os.path.exists(fileName):
print("Dataset:read_bandit_data\t[ERR]\tInput file not found at ",
fileName,
flush=True)
sys.exit(0)
f = open(fileName, 'r')
allLines = f.readlines()
f.close()
header = allLines[0]
commentIndex = header.find('#')
if commentIndex >= 0:
header = header[:commentIndex]
tokens = header.split()
numRecords = int(tokens[0])
numFeatures = int(tokens[1])
numLabels = int(tokens[2])
currIndex = 0
instanceList = []
print("Dataset:read_bandit_data\t[LOG]\tFilename: %s Number of instances: %d" %\
(fileName, numRecords), flush=True)
for i in range(numRecords):
currIndex += 1
currentLine = allLines[currIndex]
commentIndex = currentLine.find('#')
if commentIndex >= 0:
currentLine = currentLine[:commentIndex]
tokens = currentLine.split()
sampledAction = tokens[0]
sampledLoss = float(tokens[1])
sampledPropensity = float(tokens[2])
newInstance = None
sampledY = None
instanceFeature = None
if instance_type == 'MultiClass':
newInstance = Instance.MultiClass(numLabels, numFeatures)
sampledY = int(sampledAction)
instanceFeature = scipy.sparse.csr_matrix(
(1, numFeatures), dtype=numpy.longdouble)
elif instance_type == 'MultiLabel':
newInstance = Instance.MultiLabel(numLabels, numFeatures)
sampledY = numpy.zeros(numLabels, dtype=numpy.int)
if sampledAction != '-1':
for eachLabel in sampledAction.split(','):
sampledY[int(eachLabel)] = 1
instanceFeature = scipy.sparse.csr_matrix(
(1, numFeatures), dtype=numpy.longdouble)
elif instance_type == 'Brute':
newInstance = Instance.Brute(numFeatures)
sampledY = int(sampledAction)
if instance_type == 'MultiClass' or instance_type == 'MultiLabel':
for j in range(3, len(tokens)):
idVal = tokens[j].split(':')
instanceFeature[0, int(idVal[0])] = float(idVal[1])
elif instance_type == 'Brute':
numActions = int(tokens[3])
instanceFeature = scipy.sparse.csr_matrix(
(numActions, numFeatures), dtype=numpy.longdouble)
for k in range(numActions):
currIndex += 1
currentAction = allLines[currIndex]
commentIndex = currentAction.find('#')
if commentIndex >= 0:
currentAction = currentAction[:commentIndex]
tokens = currentAction.split()
currentRow = int(tokens[0])
for j in range(1, len(tokens)):
idVal = tokens[j].split(':')
instanceFeature[currentRow,
int(idVal[0])] = float(idVal[1])
newInstance.set(sampledPropensity, sampledLoss, instanceFeature,
sampledY)
instanceList.append(newInstance)
if i % 20 == 0:
print(".", flush=True, end='')
print('')
print("Dataset:read_bandit_data\t[LOG]\tFinished loading filename: %s Number of instances: %d" %\
(fileName, numRecords), flush=True)
return instanceList
