create_priv_oct.py

import numpy as np
np.seterr(all='raise')
import torch
import torchvision
import viz
import data
import latent_dp
DATA_DIR = "/mnt/ssd/data"

epsilons = [0.2, 0.5, 1.0, 2.0, 3.0, 6.0, 10.0]
sensitivity = 1.

_oct = data.chest_xray_train_data(f"{DATA_DIR}/oct")

max_s = 4.0
for epsilon in epsilons:
    save_path = f"{DATA_DIR}/priv_oct/{epsilon}"
    timestamp = "1645541874.705858"
    exp_name = f"oct_conv_cond/{timestamp}"
    priv_oct = latent_dp.privatize_pipeline(exp_name=exp_name,
                                            save_path=save_path,
                                            data_path=DATA_DIR,
                                            epsilon=epsilon,
                                            sensitivity=min(epsilon/2, max_s),
                                            mechanism="laplace")

# max_s = 4.0
# for epsilon in epsilons:
#     save_path = f"{DATA_DIR}/priv_oct_test/{epsilon}"
#     timestamp = "1645541874.705858"
#     exp_name = f"oct_conv_cond/{timestamp}"
#     priv_oct = latent_dp.privatize_pipeline(exp_name=exp_name,
#                                             save_path=save_path,
#                                             data_path=DATA_DIR,
#                                             epsilon=epsilon,
#                                             sensitivity=min(epsilon/2, max_s),
#                                             dataset="test",
#                                             mechanism="laplace")