def task(args):
import diffpri as dp
n, d, e = args
logging.info("n = %d, d = %d, e = %s", n, d, e)
if n == 0 or np.isinf(e): # no pv data -> no clipping
wx = 0.0
wy = 0.0
else:
wx, wy = dp.omega(n,d,e,mcmc)
ensure_dir_exists("drugsens_params/clipping")
with open("drugsens_params/clipping/wx_n%d_d%d_e%s.txt" % (n, d, e), 'w') as f:
f.write("%s" % wx)
with open("drugsens_params/clipping/wy_n%d_d%d_e%s.txt" % (n, d, e), 'w') as f:
f.write("%s" % wy)
if len(sys.argv) > 1:
test = int(sys.argv[1])
else:
print('No test specified as a command line argument. Choose 0 or 1.')
sys.exit()
if test == 0:
nx = len(eps)
WX = np.zeros(nx,dtype=np.float)
WY = np.zeros(nx,dtype=np.float)
for i in range(len(eps)):
e = eps[i]
n_pv = 500
n = n_pv
w_x,w_y = dp.omega(n,d,e,mcmc)
WX[i] = w_x
WY[i] = w_y
np.savetxt(csvpath+'A-WX.csv',WX,delimiter=',')
np.savetxt(csvpath+'A-WY.csv',WY,delimiter=',')
if test == 1:
nx = len(pv_size)
WX = np.zeros(nx,dtype=np.float)
WY = np.zeros(nx,dtype=np.float)
for i in range(len(pv_size)):
n_pv = pv_size[i]
n = n_pv
e = 2.0
if i == 0: # no pv data -> no clipping
WX[i] = 0.0
## NOTE: set these to match the values in tensor.py
#privacy budget: lists of similar length
eps = [1.0, 3.0, 5.0, 7.5, 10.0]
delta_list = np.zeros(shape=len(eps)) + 10e-4
np.random.seed(1)
ny = len(pv_size)
csvpath = '' # path for output csv files
privacy_for_marg_var = .3 # .3 in the paper
nx = len(eps)
WX = np.zeros((ny, nx), dtype=np.float)
WY = np.zeros((ny, nx), dtype=np.float)
print('Finding optimal projection threshold...')
for i in range(len(pv_size)):
for j in range(len(eps)):
n = pv_size[i]
d = 10
e = eps[j] * (1 - privacy_for_marg_var)
delta = delta_list[j] * (1 - privacy_for_marg_var)
w_x, w_y = dp.omega(n, d, e, delta, method='corr', ln=10)
WX[i, j] = w_x
WY[i, j] = w_y
print('WX:\n' + str(WX))
print('WY:\n' + str(WY))
print('done!')
np.savetxt(csvpath + 'C-WX.csv', WX, delimiter=',')
np.savetxt(csvpath + 'C-WY.csv', WY, delimiter=',')
# Budget split p1 = p[k] p2 = p[m] p3 = 1.0-p1-p2 # Check if split is sensible t1 = minp <= p1 and p1 <= maxp t2 = minp <= p2 and p2 <= maxp t3 = minp <= p3 and p3 <= maxp if not all([t1,t2,t3]): continue # Clipping omega and thresholds w_x,w_y = dp.omega(n,d,eps,True,ln=ln,p1=p1,p2=p2,p3=p3) c1 = sx * w_x c2 = sy * w_y # Clip data xc,yc = dp.clip(x,y,c1,c2) # Perturbed suff.stats. xx = dp.nxx(xc) + W*(c1**2.0)/p1 xy = dp.nxy(xc,yc) + L*c1*c2/p2 yy = dp.nyy(yc) + V*(c2**2.0)/p3 # Prediction pred = dp.doADVI(n,xx,xy,yy,x) # Precision