def instance(source_uri, random_seed=0):
relation_helper = RelationHelper(source_uri)
private_manager = kernel.PrivateManager(relation_helper.filename,
relation_helper.config,
random_seed=random_seed)
kernel_service = pservice.KernelService(private_manager)
return ProtectedDataSource(kernel_service,
DataManager(transformation.Null()))
def Kernel(eps_total, random_seed):
""" In an actual deployment, this code would be run on a
private server with access to the unaltered data.
"""
# Location of csv data file
filename = os.path.join(CSV_PATH, 'cps.csv')
# Configuration for data
config_file = os.path.join(CONFIG_PATH, 'cps.yml')
config = yaml.load(open(config_file, 'r').read())['cps_config']
# Private manager (or kernel) guards access to data
private_manager = kernel.PrivateManager(filename,
config,
random_seed=random_seed,
budget=eps_total)
# Kernel service mediates server-side access to kernel
kernel_service = pservice.KernelService(private_manager)
return kernel_service
CSV_PATH = os.environ['EKTELO_DATA']
CONFIG_PATH = os.path.join(os.environ['EKTELO_HOME'], 'resources', 'config')
# Setup arbitrary private constants
eps_total = 0.1
random_seed = 10
# Setup protected data source
filename = os.path.join(CSV_PATH, 'cps.csv')
config_file = os.path.join(CONFIG_PATH, 'cps.yml')
config = yaml.load(open(config_file, 'r').read())['cps_config']
private_manager = kernel.PrivateManager(filename,
config,
random_seed=random_seed,
budget=eps_total)
kernel_service = pservice.KernelService(private_manager)
R = cservice.ProtectedDataSource(kernel_service)
# Choose reduced domain for relation
domain = (10, 1, 7, 1, 1)
# Vectorize relation
x = R.vectorize(domain)
# Setup arbitrary constants for MWEM
ratio = 0.5
rounds = 3
data_scale = 1e5
use_history = True
epsilon = eps_total / 2