dist_all.py

"""
This module contains code to compute the Zest distance across 13
different pre-trained CIFAR-10 models.
"""

import os
import gc
import torch
import numpy as np

from tqdm.auto import tqdm

from zest import utils
from zest import model
from zest import train


# Experiment settings
dataset = 'CIFAR10'
base_model_path = '/home/giorgioseveri/projects/advml/lemon/cifar10_models/state_dicts/{}.pt'
all_proxies = [
    'vgg11_bn',
    'vgg13_bn',
    'vgg16_bn',
    'vgg19_bn',
    'resnet18',
    'resnet34',
    'resnet50',
    'densenet121',
    'densenet161',
    'densenet169',
    'mobilenet_v2',
    'googlenet',
    'inception_v3'
]
b_size = 128
dist = ['1', '2', 'inf', 'cos']

# Setup
all_proxies = sorted(all_proxies)
os.makedirs('results', exist_ok=True)
distances_file_name = 'distances_{}_all_{}.npy'.format(b_size, '_'.join(all_proxies))
distances_file = os.path.join('results', distances_file_name)
lime_data_name = f"{dataset}_{b_size}"

distances = {}
train_fns = {}

# Compute all LIME representations
for p in tqdm(all_proxies, desc='Computing representations'):
    p_arch = eval(f"model.{p}")
    train_fns[p] = train.TrainFn(batch_size=b_size, dataset=dataset, architecture=p_arch, lime_data_name=lime_data_name)
    train_fns[p].load(base_model_path.format(p))
    train_fns[p].lime()

# Calculate distances between all pairs
for victim_model, train_fn1 in tqdm(train_fns.items(), desc='Evaluating target model'):
    distances[victim_model] = {}

    # For each possible proxy model, load it and compute the distance
    for p in tqdm(all_proxies, desc='Computing distances'):
        if p == victim_model:
            continue

        train_fn2 = train_fns[p]
        
        distance = np.array(utils.parameter_distance(
            train_fn1.lime_mask, train_fn2.lime_mask, order=dist, lime=True))
        print('Distance between {} and {}: {}'.format(victim_model, p, distance))
        distances[victim_model][p] = distance

        del train_fn2
        gc.collect()
        torch.cuda.empty_cache()

np.save(distances_file, distances)