from utils import * from tqdm import tqdm from torch import optim from setup import setup_args from model import Network import scipy.io import os from loss import ZINBLoss if __name__ == '__main__': # for dataset_name in ["cora", "citeseer", "amap", "bat", "eat", "uat"]: for dataset_name in [ #"hrvatin_B1", #"hrvatin_B2", #"Savas" #"zhang" #"pbmc3k", #"Scala" #"Schwalbe" #"Bell" #"scp1scala" #"path1" #"path2" #"path3" #"path4" #"path5" #"path6" #"path7" #"path8" #"path9" #"path10" #"path11" #"path12" #"path13" #"path14" "path15" ]: # setup hyper-parameter args = setup_args(dataset_name) # record results file = open("result.csv", "a+") print(args.dataset, file=file) print("ACC, NMI, ARI, F1", file=file) file.close() acc_list = [] nmi_list = [] ari_list = [] f1_list = [] top_acc_values = [] top_seeds = [] top_metrics = [] # ten runs with different random seeds #seeds = [10,9] #for args.seed in seeds: for args.seed in range(args.runs): # record results # fix the random seed setup_seed(args.seed) mat_path = './data/%s.mat' % dataset_name # Load the .mat file data = scipy.io.loadmat(mat_path) # Extract 'X' and 'Y' data X = data['X'].astype(np.float32) # Ensure x is of type float32 y = data['Y'].squeeze() # Flatten y if it's multidimensional cluster_num = len(np.unique(y)) node_num = X.shape[0] A = getGraph(X.T,10) # apply the laplacian filtering X_filtered = laplacian_filtering(A, X, args.t) # test args.acc, args.nmi, args.ari, args.f1, y_hat, center = phi(X_filtered, y, cluster_num) # build our hard sample aware network scGEN = Network( input_dim=X.shape[1], hidden_dim=args.dims, act=args.activate, n_num=node_num) # adam optimizer optimizer = optim.Adam(scGEN.parameters(), lr=args.lr) # positive and negative sample pair index matrix mask = torch.ones([node_num * 2, node_num * 2]) - torch.eye(node_num * 2) # load data to device A, scGEN, X_filtered, mask = map(lambda x: x.to(args.device), (A, scGEN, X_filtered, mask)) # training for epoch in tqdm(range(args.epochs), desc="training..."): # train mode scGEN.train() # encoding with Eq. (3)-(5) Z1, Z2, E1, E2, H,mean,disp,pi = scGEN(X_filtered, A) zinb_loss_fn = ZINBLoss() zinbloss = zinb_loss_fn(H,mean,disp,pi) # calculate comprehensive similarity by Eq. (6) S = comprehensive_similarity(Z1, Z2, E1, E2, scGEN.alpha) # calculate hard sample aware contrastive loss by Eq. (10)-(11) l1 = hard_sample_aware_infoNCE(S, mask, scGEN.pos_neg_weight, scGEN.pos_weight, node_num) l2 = zinbloss if (args.Hard_only_use): loss = l1 elif (args.ZINB_only_use): loss = l2 else: loss = l1 + args.alpha*l2 # optimization loss.backward() optimizer.step() # testing and update weights of sample pairs if epoch % 10 == 0: # evaluation mode scGEN.eval() # encoding Z1, Z2, E1, E2,_,_,_,_ = scGEN(X_filtered, A) # calculate comprehensive similarity by Eq. (6) S = comprehensive_similarity(Z1, Z2, E1, E2, scGEN.alpha) # fusion and testing Z = (Z1 + Z2 + E1 + E2) / 4 acc, nmi, ari, f1, P, center = phi(Z, y, cluster_num) # select high confidence samples H, H_mat = high_confidence(Z, center) #print(H,H_mat) # calculate new weight of sample pair by Eq. (9) M, M_mat = pseudo_matrix(P, S, node_num) # update weight scGEN.pos_weight[H] = M[H].data scGEN.pos_neg_weight[H_mat] = M_mat[H_mat].data # recording if acc >= args.acc: args.acc, args.nmi, args.ari, args.f1 = acc, nmi, ari, f1 print("Training complete") # record results file = open("result.csv", "a+") print("{:.2f}, {:.2f}, {:.2f}, {:.2f}".format(args.acc, args.nmi, args.ari, args.f1), file=file) file.close() acc_list.append(args.acc) nmi_list.append(args.nmi) ari_list.append(args.ari) f1_list.append(args.f1) # Store the top 2 ACC values and seeds if len(top_acc_values) < 2: top_acc_values.append(args.acc) top_seeds.append(args.seed) top_metrics.append((args.acc, args.nmi, args.ari, args.f1)) else: min_index = top_acc_values.index(min(top_acc_values)) if args.acc > top_acc_values[min_index]: top_acc_values[min_index] = args.acc top_seeds[min_index] = args.seed top_metrics[min_index] = (args.acc, args.nmi, args.ari, args.f1) # Print the top 2 ACC values and seeds combined = list(zip(top_acc_values, top_seeds, top_metrics)) combined.sort(key=lambda x: x[0], reverse=True) # 按ACC降序排序 # 解包排序后的结果 top_acc_values = [x[0] for x in combined] top_seeds = [x[1] for x in combined] top_metrics = [x[2] for x in combined] for i in range(2): print(f"Top {i + 1} ACC: {top_acc_values[i]:.2f} on run {top_seeds[i]}") print( f"Metrics: ACC={top_metrics[i][0]:.2f}, NMI={top_metrics[i][1]:.2f}, ARI={top_metrics[i][2]:.2f}, F1={top_metrics[i][3]:.2f}") # Final recording for the top 2 ACC values and metrics with open("result.csv", "a+") as file: for i in range(2): file.write( f"{top_seeds[i]},{top_metrics[i][0]:.2f},{top_metrics[i][1]:.2f},{top_metrics[i][2]:.2f},{top_metrics[i][3]:.2f}\n") # Convert top_metrics to a NumPy array for easy statistics calculation top_metrics_array = np.array(top_metrics) mean_metrics = top_metrics_array.mean(axis=0) std_metrics = top_metrics_array.std(axis=0) # Print and record the mean and std of the top metrics with open("result.csv", "a+") as file: file.write(f"{mean_metrics[0]:.2f},{std_metrics[0]:.2f}\n") file.write(f"{mean_metrics[1]:.2f},{std_metrics[1]:.2f}\n") file.write(f"{mean_metrics[2]:.2f},{std_metrics[2]:.2f}\n") file.write(f"{mean_metrics[3]:.2f},{std_metrics[3]:.2f}\n")