if __name__ == "__main__":
parser = argparse.ArgumentParser()
# ol: target classifier is from the adapttive attack
# kkt: target is from kkt attack
# real: actual classifier, compare: compare performance
# of kkt attack and adaptive attack using same stop criteria
# Global params
parser.add_argument('--model_arch',
default='lenet',
choices=dnn_utils.get_model_names(),
help='Victim model architecture')
parser.add_argument('--dataset',
default='mnist',
choices=datasets.get_dataset_names(),
help="Which dataset to use?")
parser.add_argument('--batch_size',
default=-1,
type=int,
help='Batch size while training models')
parser.add_argument('--online_alg_criteria',
default='norm',
choices=['max_loss', 'norm'],
help='Stop criteria of online alg: max_loss or norm')
parser.add_argument('--poison_model_path',
type=str,
help='Path to saved poisoned-classifier')
parser.add_argument('--log_path',
type=str,
default="./data/logs",
import math
import argparse
import numpy as np
import torch
import torch.utils.data
import torchvision.transforms as transforms
from datasets import create_dataset, get_dataset_names
parser = argparse.ArgumentParser(description='Dataset statistics calculator')
parser.add_argument('data', metavar='DIR', help='path to dataset')
parser.add_argument('--dataset',
default='tum',
help='dataset type: ' + ' | '.join(get_dataset_names()) +
' (default: tum)')
parser.add_argument('--input-channels',
default=3,
type=int,
dest='input_channels')
parser.add_argument('-b', '--batch-size', default=128, type=int)
parser.add_argument('-j', '--workers', default=8, type=int)
args = parser.parse_args()
# load the dataset
dataset = create_dataset(args.dataset,
root_dir=args.data,
type='train',
input_channels=args.input_channels,
import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf
import tensorflow_probability as tfp
from tqdm import tqdm
import utils as utl
import models as m
import datasets as d
tfd = tfp.distributions
tfm = tf.math
# Set root of where data is
d.root = '../../datasets/raw/'
#%% Load data
dataset_names = d.get_dataset_names()
name = dataset_names[5]
data, X_train, X_val, X_test = d.load_data(name)
M = X_train.shape[1]
print(f'\nX_train.shape = {X_train.shape}')
print(f'\nX_val.shape = {X_val.shape}')
print(f'\nX_train.shape = {X_train.shape}')
print(name + ' data loaded...')
#%% Parameters
model_name = 'TT'
epochs = 500
N_init = 5 # Number of random initializations to do
batch_size = 1000
import torch
import torch.nn as nn
import torch.nn.parallel
import torch.backends.cudnn as cudnn
import torch.distributed as dist
import torch.optim
import torch.multiprocessing as mp
import torch.utils.data
import torch.utils.data.distributed
import torchvision.transforms as transforms
from models import create_model, get_model_names
from datasets import create_dataset, get_dataset_names, DataLoaderCache
model_names = get_model_names()
dataset_names = get_dataset_names()
#
# parse command-line arguments
#
parser = argparse.ArgumentParser(description='PyTorch OdometryNet Training')
parser.add_argument('data', metavar='DIR', help='path to dataset')
parser.add_argument('--dataset',
default='tum',
help='dataset type: ' + ' | '.join(dataset_names) +
' (default: tum)')
parser.add_argument('--model-dir',
type=str,
default='',
help='path to desired output directory for saving model '