-
Notifications
You must be signed in to change notification settings - Fork 3
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
- Loading branch information
Showing
6 changed files
with
252 additions
and
31 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,134 @@ | ||
| import numpy as np | ||
| import random | ||
| import torch | ||
| import copy | ||
|
|
||
| from explainers.GNNExplainer.torch_geom_our.out import GNNExplainer | ||
| from explainers.SubgraphX.out import SubgraphXExplainer | ||
| from explainers.Zorro.out import ZorroExplainer | ||
| from aux.utils import EXPLAINERS_INIT_PARAMETERS_PATH, EXPLAINERS_LOCAL_RUN_PARAMETERS_PATH, \ | ||
| EXPLAINERS_GLOBAL_RUN_PARAMETERS_PATH | ||
| from aux.configs import ConfigPattern | ||
|
|
||
| from tqdm import tqdm | ||
| from networkx.classes import neighbors | ||
| from numpy.array_api import astype | ||
| from sympy.codegen.ast import int64 | ||
|
|
||
| from attacks.evasion_attacks import EvasionAttacker | ||
| from aux.configs import CONFIG_OBJ | ||
| from explainers.explainer import ProgressBar | ||
| from typing import Dict, Optional | ||
|
|
||
|
|
||
| class EAttack(EvasionAttacker): | ||
| name = "EAttack" | ||
|
|
||
| def __init__(self, explainer, run_config, attack_size, attack_inds, targeted, max_rewire, random_rewire, | ||
| attack_edges, attack_features, edge_mode, features_mode, **kwargs): | ||
| super().__init__(**kwargs) | ||
| self.explainer = explainer | ||
| self.run_config = run_config | ||
| # self.mode = mode | ||
| self.mode = getattr(run_config, CONFIG_OBJ).mode | ||
| self.params = getattr(getattr(run_config, CONFIG_OBJ).kwargs, CONFIG_OBJ).to_dict() | ||
| self.attack_size = attack_size | ||
| self.targeted = targeted | ||
| self.max_rewire = max_rewire | ||
| self.attack_inds = attack_inds | ||
| self.random_rewire = random_rewire | ||
| self.attack_edges = attack_edges | ||
| self.attack_features = attack_features | ||
| self.edge_mode = edge_mode | ||
| self.features_mode = features_mode | ||
|
|
||
|
|
||
| def attack(self, model_manager, gen_dataset, mask_tensor): | ||
|
|
||
| assert self.attack_edges or self.attack_features | ||
|
|
||
| explanations = [] | ||
|
|
||
| # get explanations | ||
| for i in tqdm(range(len(self.attack_inds))): | ||
| self.params['element_idx'] = self.attack_inds[i] | ||
| self.explainer.pbar = ProgressBar(None, "er", desc=f'{self.explainer.name} explaining') | ||
| self.explainer.run(self.mode, self.params, finalize=True) | ||
| explanation = copy.deepcopy(self.explainer.explanation.dictionary['data']) | ||
| explanations.append(explanation) | ||
|
|
||
| edge_index = gen_dataset.dataset.data.edge_index.tolist() | ||
| edge_index_set = set([(u, v) for u, v in zip(edge_index[0], edge_index[1])]) | ||
|
|
||
| if self.attack_edges: | ||
| cnt = 0 | ||
|
|
||
| for i, n in enumerate(self.attack_inds): | ||
| if self.edge_mode == 'remove': | ||
| for e in explanations[i]['edges'].keys(): | ||
| u, v = map(int, e.split(',')) | ||
| if u != n and v != n: # not remove within 1-hop | ||
| # TODO check with discard of (v, u) too | ||
| cnt += 1 | ||
| edge_index_set.discard((u, v)) | ||
| # TEST | ||
| edge_index_set.discard((v, u)) | ||
| elif self.edge_mode == 'add': | ||
| unimportant_nodes = set() | ||
| important_nodes = set() | ||
| for (u, v) in zip(edge_index[0], edge_index[1]): | ||
| if v == n: | ||
| if f"{u},{v}" not in explanations[i]['edges'].keys(): | ||
| unimportant_nodes.add(u) | ||
| else: | ||
| important_nodes.add(u) | ||
| elif u == n: | ||
| if f"{u},{v}" not in explanations[i]['edges'].keys(): | ||
| unimportant_nodes.add(v) | ||
| else: | ||
| important_nodes.add(v) | ||
| # if ((v == n and f"{u},{v}" not in explanations[i]['edges'].keys()) and | ||
| # f"{v},{u}" not in explanations[i]['edges'].keys()): | ||
| # unimportant_nodes.add(u) | ||
| # elif v == n: | ||
| # important_nodes.add(u) | ||
| unimportant_nodes = list(unimportant_nodes) | ||
| # TEST | ||
| edges = [(u, v) for u, v in zip(edge_index[0], edge_index[1]) if u == n or v == n] | ||
| #print(len(edges)) | ||
| if len(unimportant_nodes) == 0: | ||
| continue | ||
| for e in explanations[i]['edges'].keys(): | ||
| u, v = map(int, e.split(',')) | ||
| if v in important_nodes and u != n: | ||
| new_node = random.sample(unimportant_nodes, 1) | ||
| edge_index_set.add((u, new_node[0])) | ||
| cnt += 1 | ||
| elif self.edge_mode == 'rewire': | ||
| for (u, v) in zip(edge_index[0], edge_index[1]): | ||
| if u != n and v != n and f"{u},{v}" in explanations[i]['edges'].keys(): | ||
| edge_index_set.discard((u, v)) | ||
| if (u, n) not in edge_index_set: | ||
| cnt += 1 | ||
| edge_index_set.add((u, n)) | ||
| elif (v, n) not in edge_index_set: | ||
| cnt += 1 | ||
| edge_index_set.add((v, n)) | ||
|
|
||
| # Update dataset edges | ||
| edge_index_new = [[], []] | ||
| for (u, v) in edge_index_set: | ||
| edge_index_new[0].append(u) | ||
| edge_index_new[1].append(v) | ||
| edge_index_new = torch.tensor(edge_index_new, dtype=torch.int64) | ||
| gen_dataset.dataset.data.edge_index = edge_index_new | ||
|
|
||
| print(cnt) | ||
|
|
||
| if self.attack_features: | ||
| if self.features_mode == 'reverse': | ||
| pass | ||
| elif self.features_mode == 'drop': | ||
| pass | ||
|
|
||
| return gen_dataset |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters