import torch import numpy as np import torch.nn.functional as F import scipy.sparse as sp from deeprobust.graph.defense import GCNJaccard, GCN, GAT from deeprobust.graph.utils import * from deeprobust.graph.data import Dataset, PrePtbDataset, Dpr2Pyg import argparse import csv import os parser = argparse.ArgumentParser() parser.add_argument('--seed', type=int, default=15, help='Random seed.') parser.add_argument('--dataset', type=str, default='pubmed', choices=['cora', 'cora_ml', 'citeseer', 'polblogs', 'pubmed', 'Flickr'], help='dataset') parser.add_argument('--ptb_rate', type=float, default=0.25, help='pertubation rate') parser.add_argument('--ptb_type', type=str, default='dice', choices=['clean', 'meta', 'dice', 'minmax', 'pgd', 'random'], help='attack type') parser.add_argument('--hidden', type=int, default=16, help='Number of hidden units.') parser.add_argument('--dropout', type=float, default=0.5, help='Dropout rate (1 - keep probability).') parser.add_argument('--gpu', type=int, default=0, help='GPU device ID (default: 0)') args = parser.parse_args() args.cuda = torch.cuda.is_available() print('cuda: %s' % args.cuda) device = torch.device(f"cuda:{args.gpu}" if torch.cuda.is_available() else "cpu") # make sure you use the same data splits as you generated attacks np.random.seed(args.seed) if args.cuda: torch.cuda.manual_seed(args.seed) # Here the random seed is to split the train/val/test data, # we need to set the random seed to be the same as that when you generate the perturbed graph # data = Dataset(root='/tmp/', name=args.dataset, setting='nettack', seed=15) # Or we can just use setting='prognn' to get the splits # data = Dataset(root='/tmp/', name=args.dataset, setting='prognn') data = Dataset(root='/tmp/', name=args.dataset) adj, features, labels = data.adj, data.features, data.labels idx_train, idx_val, idx_test = data.idx_train, data.idx_val, data.idx_test idx_unlabeled = np.union1d(idx_val, idx_test) # print(type((adj))) # adj, features, labels = preprocess(adj, features, labels, preprocess_adj=False) ## load from attacked_adj ptb_path = f"../attacked_adj/{args.dataset}/{args.ptb_type}_{args.dataset}_{args.ptb_rate}.pt" perturbed_adj = torch.load(ptb_path) print("type(perturbed_adj)", type(perturbed_adj)) perturbed_adj = sp.csr_matrix(perturbed_adj.cpu().numpy()) gat = GAT(nfeat=features.shape[1], nhid=8, heads=8, nclass=labels.max().item() + 1, dropout=0.5, device=device) gat = gat.to(device) # test on clean graph print('==================') print('=== train on clean graph ===') pyg_data = Dpr2Pyg(data) pyg_data.update_edge_index(perturbed_adj) # inplace operation gat.fit(pyg_data, train_iters=500, verbose=True) # train with earlystopping acc = gat.test() print("acc:", acc) csv_dir = "../result" os.makedirs(csv_dir, exist_ok=True) csv_filename = os.path.join(csv_dir, f"GAT_{args.dataset}_{args.ptb_type}_{args.ptb_rate}.csv") row = [f"{args.dataset} ", f" {args.ptb_type} ", f" {args.ptb_rate} ", f" {acc}"] try: file_exists = os.path.isfile(csv_filename) with open(csv_filename, 'a', newline='') as csvfile: writer = csv.writer(csvfile) if not file_exists: writer.writerow(["dataset ", "ptb_type ", "ptb_rate ", "accuracy"]) writer.writerow(row) except Exception as e: print(f"[Error] Failed to write CSV: {e}")