def RapporClientSim(params, irr_rand, csv_in):
"""Read true values from csv_in and output encoded values on csv_out."""
header = ['client', 'cohort', 'bloom', 'prr', 'irr']
out_rows = {}
# TODO: It would be more instructive/efficient to construct an encoder
# instance up front per client, rather than one per row below.
for i, (index, client_str, cohort_str, true_value) in csv_in.iterrows():
cohort = int(cohort_str)
secret = client_str
e = rappor.Encoder(params, cohort, secret, irr_rand)
# Real users should call e.encode(). For testing purposes, we also want
# the PRR.
bloom, prr, irr = e._internal_encode(true_value)
bloom_str = rappor.bit_string(bloom, params.num_bloombits)
prr_str = rappor.bit_string(prr, params.num_bloombits)
irr_str = rappor.bit_string(irr, params.num_bloombits)
out_rows[i] = [client_str, cohort_str, bloom_str, prr_str, irr_str]
output = pd.DataFrame.from_dict(out_rows, orient='index')
output.columns = header
output.to_csv('/Users/Michael/PycharmProjects/untitled1/data/output.csv')
def GenAssocTestdata(params1, params2, irr_rand, assoc_testdata_count, csv_in,
csv_out):
"""Read true values from csv_in and output encoded values on csv_out.
Replicate assoc_testdata_count times. First value is a string, second is a
bool. TODO: Generalize this.
"""
rows = []
for i, (true_value1, true_value2) in enumerate(csv_in):
if i == 0:
v1_name = true_value1
v2_name = true_value2
continue # skip header row
rows.append((true_value1, true_value2))
# Use the same column names
header = ('client', 'cohort', v1_name, v2_name)
csv_out.writerow(header)
n = assoc_testdata_count
report_index = 0
for i in xrange(n):
for v1, v2 in rows:
client_str = 'c%d' % report_index
# randint(a, b) gives i such that a <= i <= b
cohort = random.randint(0, params1.num_cohorts - 1)
string_encoder = rappor.Encoder(params1, cohort, client_str,
irr_rand)
bool_encoder = rappor.Encoder(params2, cohort, client_str,
irr_rand)
# Real users should call e.encode(). For testing purposes, we also want
# the PRR.
irr1 = string_encoder.encode(v1)
# TODO: Convert to bool and encode with basic RAPPOR
v2_int = int(v2)
#print v2_int
irr2 = bool_encoder.encode_bits(v2_int)
irr1_str = rappor.bit_string(irr1, params1.num_bloombits)
irr2_str = rappor.bit_string(irr2, params2.num_bloombits)
csv_out.writerow((client_str, cohort, irr1_str, irr2_str))
report_index += 1
def GenAssocTestdata(params1, params2, irr_rand, assoc_testdata_count,
csv_in, csv_out):
"""Read true values from csv_in and output encoded values on csv_out.
Replicate assoc_testdata_count times. First value is a string, second is a
bool. TODO: Generalize this.
"""
rows = []
for i, (true_value1, true_value2) in enumerate(csv_in):
if i == 0:
v1_name = true_value1
v2_name = true_value2
continue # skip header row
rows.append((true_value1, true_value2))
# Use the same column names
header = ('client', 'cohort', v1_name, v2_name)
csv_out.writerow(header)
n = assoc_testdata_count
report_index = 0
for i in xrange(n):
for v1, v2 in rows:
client_str = 'c%d' % report_index
# randint(a, b) gives i such that a <= i <= b
cohort = random.randint(0, params1.num_cohorts - 1)
string_encoder = rappor.Encoder(params1, cohort, client_str, irr_rand)
bool_encoder = rappor.Encoder(params2, cohort, client_str, irr_rand)
# Real users should call e.encode(). For testing purposes, we also want
# the PRR.
irr1 = string_encoder.encode(v1)
# TODO: Convert to bool and encode with basic RAPPOR
v2_int = int(v2)
#print v2_int
irr2 = bool_encoder.encode_bits(v2_int)
irr1_str = rappor.bit_string(irr1, params1.num_bloombits)
irr2_str = rappor.bit_string(irr2, params2.num_bloombits)
csv_out.writerow((client_str, cohort, irr1_str, irr2_str))
report_index += 1
def RapporClientSim(params, irr_rand, csv_in, csv_out):
"""Read true values from csv_in and output encoded values on csv_out."""
header = ('client', 'cohort', 'bloom', 'prr', 'irr')
csv_out.writerow(header)
# TODO: It would be more instructive/efficient to construct an encoder
# instance up front per client, rather than one per row below.
start_time = time.time()
for i, (client_str, cohort_str, true_value) in enumerate(csv_in):
if i == 0:
if client_str != 'client':
raise RuntimeError('Expected client header, got %s' %
client_str)
if cohort_str != 'cohort':
raise RuntimeError('Expected cohort header, got %s' %
cohort_str)
if true_value != 'value':
raise RuntimeError('Expected value header, got %s' % value)
continue # skip header row
#if i == 30: # EARLY STOP
# break
if i % 10000 == 0:
elapsed = time.time() - start_time
log('Processed %d inputs in %.2f seconds', i, elapsed)
cohort = int(cohort_str)
secret = client_str
e = rappor.Encoder(params, cohort, secret, irr_rand)
# Real users should call e.encode(). For testing purposes, we also want
# the PRR.
bloom, prr, irr = e._internal_encode(true_value)
bloom_str = rappor.bit_string(bloom, params.num_bloombits)
prr_str = rappor.bit_string(prr, params.num_bloombits)
irr_str = rappor.bit_string(irr, params.num_bloombits)
out_row = (client_str, cohort_str, bloom_str, prr_str, irr_str)
csv_out.writerow(out_row)
def RapporClientSim(params, irr_rand, csv_in, csv_out):
"""Read true values from csv_in and output encoded values on csv_out."""
header = ('client', 'cohort', 'bloom', 'prr', 'irr')
csv_out.writerow(header)
# TODO: It would be more instructive/efficient to construct an encoder
# instance up front per client, rather than one per row below.
start_time = time.time()
for i, (client_str, cohort_str, true_value) in enumerate(csv_in):
if i == 0:
if client_str != 'client':
raise RuntimeError('Expected client header, got %s' % client_str)
if cohort_str != 'cohort':
raise RuntimeError('Expected cohort header, got %s' % cohort_str)
if true_value != 'value':
raise RuntimeError('Expected value header, got %s' % value)
continue # skip header row
#if i == 30: # EARLY STOP
# break
if i % 10000 == 0:
elapsed = time.time() - start_time
log('Processed %d inputs in %.2f seconds', i, elapsed)
cohort = int(cohort_str)
secret = client_str
e = rappor.Encoder(params, cohort, secret, irr_rand)
# Real users should call e.encode(). For testing purposes, we also want
# the PRR.
bloom, prr, irr = e._internal_encode(true_value)
bloom_str = rappor.bit_string(bloom, params.num_bloombits)
prr_str = rappor.bit_string(prr, params.num_bloombits)
irr_str = rappor.bit_string(irr, params.num_bloombits)
out_row = (client_str, cohort_str, bloom_str, prr_str, irr_str)
csv_out.writerow(out_row)
def main(argv):
(opts, argv) = CreateOptionsParser().parse_args(argv)
# Copy flags into params
params = rappor.Params()
params.num_bloombits = opts.num_bits
params.num_hashes = opts.num_hashes
params.num_cohorts = opts.num_cohorts
params.prob_p = opts.prob_p
params.prob_q = opts.prob_q
params.prob_f = opts.prob_f
if opts.random_mode == 'simple':
irr_rand = rappor.SimpleIrrRand(params)
elif opts.random_mode == 'fast':
if fastrand:
log('Using fastrand extension')
# NOTE: This doesn't take 'rand'. It's seeded in C with srand().
irr_rand = fastrand.FastIrrRand(params)
else:
log('Warning: fastrand module not importable; see README for build '
'instructions. Falling back to simple randomness.')
irr_rand = rappor.SimpleIrrRand(params)
else:
raise AssertionError
# Other possible implementations:
# - random.SystemRandom (probably uses /dev/urandom on Linux)
# - HMAC-SHA256 with another secret? This could match C++ byte for byte.
# - or srand(0) might do it.
csv_in = csv.reader(sys.stdin)
csv_out = csv.writer(sys.stdout)
header = ('client', 'cohort', 'bloom', 'prr', 'irr')
csv_out.writerow(header)
# TODO: It would be more instructive/efficient to construct an encoder
# instance up front per client, rather than one per row below.
start_time = time.time()
for i, (client_str, cohort_str, true_value) in enumerate(csv_in):
if i == 0:
if client_str != 'client':
raise RuntimeError('Expected client header, got %s' % client_str)
if cohort_str != 'cohort':
raise RuntimeError('Expected cohort header, got %s' % cohort_str)
if true_value != 'value':
raise RuntimeError('Expected value header, got %s' % value)
continue # skip header row
#if i == 30: # EARLY STOP
# break
if i % 10000 == 0:
elapsed = time.time() - start_time
log('Processed %d inputs in %.2f seconds', i, elapsed)
cohort = int(cohort_str)
secret = client_str
e = rappor.Encoder(params, cohort, secret, irr_rand)
# Real users should call e.encode(). For testing purposes, we also want
# the PRR.
bloom, prr, irr = e._internal_encode(true_value)
bloom_str = rappor.bit_string(bloom, params.num_bloombits)
prr_str = rappor.bit_string(prr, params.num_bloombits)
irr_str = rappor.bit_string(irr, params.num_bloombits)
out_row = (client_str, cohort_str, bloom_str, prr_str, irr_str)
csv_out.writerow(out_row)