def train_val_pipeline(MODEL_NAME, DATASET_NAME, params, net_params, dirs): avg_test_acc = [] avg_train_acc = [] avg_epochs = [] t0 = time.time() per_epoch_time = [] dataset = LoadData(DATASET_NAME) if MODEL_NAME in ['GCN', 'GAT']: if net_params['self_loop']: print( "[!] Adding graph self-loops for GCN/GAT models (central node trick)." ) dataset._add_self_loops() if net_params['pos_enc']: print("[!] Adding graph positional encoding.") dataset._add_positional_encodings(net_params['pos_enc_dim']) #TODO trainset, valset, testset = dataset.train, dataset.val, dataset.test root_log_dir, root_ckpt_dir, write_file_name, write_config_file = dirs device = net_params['device'] # Write the network and optimization hyper-parameters in folder config/ with open(write_config_file + '.txt', 'w') as f: f.write("""Dataset: {},\nModel: {}\n\nparams={}\n\nnet_params={}\n\n\nTotal Parameters: {}\n\n""" \ .format(DATASET_NAME, MODEL_NAME, params, net_params, net_params['total_param'])) # At any point you can hit Ctrl + C to break out of training early. try: for split_number in range(5): t0_split = time.time() log_dir = os.path.join(root_log_dir, "RUN_" + str(split_number)) writer = SummaryWriter(log_dir=log_dir) # setting seeds random.seed(params['seed']) np.random.seed(params['seed']) torch.manual_seed(params['seed']) if device.type == 'cuda': torch.cuda.manual_seed(params['seed']) print("RUN NUMBER: ", split_number) trainset, valset, testset = dataset.train[ split_number], dataset.val[split_number], dataset.test[ split_number] print("Training Graphs: ", len(trainset)) print("Validation Graphs: ", len(valset)) print("Test Graphs: ", len(testset)) print("Number of Classes: ", net_params['n_classes']) model = gnn_model(MODEL_NAME, net_params) model = model.to(device) optimizer = optim.Adam(model.parameters(), lr=params['init_lr'], weight_decay=params['weight_decay']) scheduler = optim.lr_scheduler.ReduceLROnPlateau( optimizer, mode='min', factor=params['lr_reduce_factor'], patience=params['lr_schedule_patience'], verbose=True) epoch_train_losses, epoch_val_losses = [], [] epoch_train_accs, epoch_val_accs = [], [] # batching exception for Diffpool drop_last = True if MODEL_NAME == 'DiffPool' else False # drop_last = False if MODEL_NAME in ['RingGNN', '3WLGNN']: # import train functions specific for WL-GNNs from train.train_CSL_graph_classification import train_epoch_dense as train_epoch, evaluate_network_dense as evaluate_network from functools import partial # util function to pass pos_enc flag to collate function train_loader = DataLoader(trainset, shuffle=True, collate_fn=partial( dataset.collate_dense_gnn, pos_enc=net_params['pos_enc'])) val_loader = DataLoader(valset, shuffle=False, collate_fn=partial( dataset.collate_dense_gnn, pos_enc=net_params['pos_enc'])) test_loader = DataLoader(testset, shuffle=False, collate_fn=partial( dataset.collate_dense_gnn, pos_enc=net_params['pos_enc'])) else: # import train functions for all other GCNs from train.train_CSL_graph_classification import train_epoch_sparse as train_epoch, evaluate_network_sparse as evaluate_network train_loader = DataLoader(trainset, batch_size=params['batch_size'], shuffle=True, drop_last=drop_last, collate_fn=dataset.collate) val_loader = DataLoader(valset, batch_size=params['batch_size'], shuffle=False, drop_last=drop_last, collate_fn=dataset.collate) test_loader = DataLoader(testset, batch_size=params['batch_size'], shuffle=False, drop_last=drop_last, collate_fn=dataset.collate) with tqdm(range(params['epochs'])) as t: for epoch in t: t.set_description('Epoch %d' % epoch) start = time.time() if MODEL_NAME in [ 'RingGNN', '3WLGNN' ]: # since different batch training function for dense GNNs epoch_train_loss, epoch_train_acc, optimizer = train_epoch( model, optimizer, device, train_loader, epoch, params['batch_size']) else: # for all other models common train function epoch_train_loss, epoch_train_acc, optimizer = train_epoch( model, optimizer, device, train_loader, epoch) # epoch_train_loss, epoch_train_acc, optimizer = train_epoch(model, optimizer, device, train_loader, epoch) epoch_val_loss, epoch_val_acc = evaluate_network( model, device, val_loader, epoch) _, epoch_test_acc = evaluate_network( model, device, test_loader, epoch) epoch_train_losses.append(epoch_train_loss) epoch_val_losses.append(epoch_val_loss) epoch_train_accs.append(epoch_train_acc) epoch_val_accs.append(epoch_val_acc) writer.add_scalar('train/_loss', epoch_train_loss, epoch) writer.add_scalar('val/_loss', epoch_val_loss, epoch) writer.add_scalar('train/_acc', epoch_train_acc, epoch) writer.add_scalar('val/_acc', epoch_val_acc, epoch) writer.add_scalar('test/_acc', epoch_test_acc, epoch) writer.add_scalar('learning_rate', optimizer.param_groups[0]['lr'], epoch) epoch_train_acc = 100. * epoch_train_acc epoch_test_acc = 100. * epoch_test_acc t.set_postfix(time=time.time() - start, lr=optimizer.param_groups[0]['lr'], train_loss=epoch_train_loss, val_loss=epoch_val_loss, train_acc=epoch_train_acc, val_acc=epoch_val_acc, test_acc=epoch_test_acc) per_epoch_time.append(time.time() - start) # Saving checkpoint ckpt_dir = os.path.join(root_ckpt_dir, "RUN_" + str(split_number)) if not os.path.exists(ckpt_dir): os.makedirs(ckpt_dir) torch.save( model.state_dict(), '{}.pkl'.format(ckpt_dir + "/epoch_" + str(epoch))) files = glob.glob(ckpt_dir + '/*.pkl') for file in files: epoch_nb = file.split('_')[-1] epoch_nb = int(epoch_nb.split('.')[0]) if epoch_nb < epoch - 1: os.remove(file) scheduler.step(epoch_val_loss) if optimizer.param_groups[0]['lr'] < params['min_lr']: print("\n!! LR EQUAL TO MIN LR SET.") break # Stop training after params['max_time'] hours if time.time() - t0_split > params[ 'max_time'] * 3600 / 10: # Dividing max_time by 10, since there are 10 runs in TUs print('-' * 89) print( "Max_time for one train-val-test split experiment elapsed {:.3f} hours, so stopping" .format(params['max_time'] / 10)) break _, test_acc = evaluate_network(model, device, test_loader, epoch) _, train_acc = evaluate_network(model, device, train_loader, epoch) avg_test_acc.append(test_acc) avg_train_acc.append(train_acc) avg_epochs.append(epoch) print("Test Accuracy [LAST EPOCH]: {:.4f}".format(test_acc)) print("Train Accuracy [LAST EPOCH]: {:.4f}".format(train_acc)) except KeyboardInterrupt: print('-' * 89) print('Exiting from training early because of KeyboardInterrupt') print("TOTAL TIME TAKEN: {:.4f}hrs".format((time.time() - t0) / 3600)) print("AVG TIME PER EPOCH: {:.4f}s".format(np.mean(per_epoch_time))) # Final test accuracy value averaged over 5-fold print("""\n\n\nFINAL RESULTS\n\nTEST ACCURACY averaged: {:.4f} with s.d. {:.4f}""" \ .format(np.mean(np.array(avg_test_acc)) * 100, np.std(avg_test_acc) * 100)) print("\nAll splits Test Accuracies:\n", avg_test_acc) print("""\n\n\nFINAL RESULTS\n\nTRAIN ACCURACY averaged: {:.4f} with s.d. {:.4f}""" \ .format(np.mean(np.array(avg_train_acc)) * 100, np.std(avg_train_acc) * 100)) print("\nAll splits Train Accuracies:\n", avg_train_acc) writer.close() """ Write the results in out/results folder """ with open(write_file_name + '.txt', 'w') as f: f.write("""Dataset: {},\nModel: {}\n\nparams={}\n\nnet_params={}\n\n{}\n\nTotal Parameters: {}\n\n FINAL RESULTS\nTEST ACCURACY averaged: {:.3f}\n with test acc s.d. {:.3f}\nTRAIN ACCURACY averaged: {:.3f}\n with train s.d. {:.3f}\n\n Convergence Time (Epochs): {:.3f}\nTotal Time Taken: {:.3f} hrs\nAverage Time Per Epoch: {:.3f} s\n\n\nAll Splits Test Accuracies: {}\n\nAll Splits Train Accuracies: {}""" \ .format(DATASET_NAME, MODEL_NAME, params, net_params, model, net_params['total_param'], np.mean(np.array(avg_test_acc)) * 100, np.std(avg_test_acc) * 100, np.mean(np.array(avg_train_acc)) * 100, np.std(avg_train_acc) * 100, np.mean(np.array(avg_epochs)), (time.time() - t0) / 3600, np.mean(per_epoch_time), avg_test_acc, avg_train_acc))
def train_val_pipeline(MODEL_NAME, DATASET_NAME, params, net_params, args): # setting seeds random.seed(params['seed']) np.random.seed(params['seed']) torch.manual_seed(params['seed']) device = net_params['device'] if device == 'cuda': torch.cuda.manual_seed(params['seed']) dataset = LoadData(DATASET_NAME) trainset, valset, testset = dataset.train, dataset.val, dataset.test net_params['in_dim'] = torch.unique(dataset.train[0][0].ndata['feat'], dim=0).size( 0) # node_dim (feat is an integer) net_params['n_classes'] = torch.unique(dataset.train[0][1], dim=0).size(0) net_params['total_param'] = view_model_param(MODEL_NAME, net_params) load_model = args.load_model aug_type_list = [ 'drop_nodes', 'drop_add_edges', 'noise', 'mask', 'subgraph' ] if MODEL_NAME in ['GCN', 'GAT']: if net_params['self_loop']: print( "[!] Adding graph self-loops for GCN/GAT models (central node trick)." ) dataset._add_self_loops() print('-' * 40 + "Finetune Option" + '-' * 40) print('SEED: [{}]'.format(params['seed'])) print("Data Name: [{}]".format(DATASET_NAME)) print("Model Name: [{}]".format(MODEL_NAME)) print("Training Graphs:[{}]".format(len(trainset))) print("Valid Graphs: [{}]".format(len(valset))) print("Test Graphs: [{}]".format(len(testset))) print("Number Classes: [{}]".format(net_params['n_classes'])) print("Learning rate: [{}]".format(params['init_lr'])) print('-' * 40 + "Contrastive Option" + '-' * 40) print("Load model: [{}]".format(load_model)) print("Aug Type: [{}]".format(aug_type_list[args.aug])) print("Projection head:[{}]".format(args.head)) print('-' * 100) model = gnn_model(MODEL_NAME, net_params) if load_model: output_path = './001_contrastive_models' save_model_dir0 = os.path.join(output_path, DATASET_NAME) save_model_dir1 = os.path.join(save_model_dir0, aug_type_list[args.aug]) if args.head: save_model_dir1 += "_head" else: save_model_dir1 += "_no_head" save_model_dir2 = os.path.join(save_model_dir1, MODEL_NAME) load_file_name = glob.glob(save_model_dir2 + '/*.pkl') checkpoint = torch.load(load_file_name[-1]) model_dict = model.state_dict() state_dict = { k: v for k, v in checkpoint.items() if k in model_dict.keys() } model.load_state_dict(state_dict) print('Success load pre-trained model!: [{}]'.format( load_file_name[-1])) else: print('No model load!: Test baseline! ') model = model.to(device) optimizer = optim.Adam(model.parameters(), lr=params['init_lr'], weight_decay=params['weight_decay']) scheduler = optim.lr_scheduler.ReduceLROnPlateau( optimizer, mode='min', factor=params['lr_reduce_factor'], patience=params['lr_schedule_patience'], verbose=True) train_loader = DataLoader(trainset, batch_size=params['batch_size'], shuffle=True, collate_fn=dataset.collate) val_loader = DataLoader(valset, batch_size=params['batch_size'], shuffle=False, collate_fn=dataset.collate) test_loader = DataLoader(testset, batch_size=params['batch_size'], shuffle=False, collate_fn=dataset.collate) for epoch in range(params['epochs']): epoch_train_loss, epoch_train_acc, optimizer = train_epoch( model, optimizer, device, train_loader, epoch) epoch_val_loss, epoch_val_acc = evaluate_network( model, device, val_loader, epoch) epoch_test_loss, epoch_test_acc = evaluate_network( model, device, test_loader, epoch) _, epoch_test_acc = evaluate_network(model, device, test_loader, epoch) print('-' * 80) print( time.strftime("%Y-%m-%d %H:%M:%S", time.localtime()) + ' | ' + "Epoch [{:>2d}] Test Acc: [{:.4f}]".format( epoch + 1, epoch_test_acc)) print('-' * 80) scheduler.step(epoch_val_loss) if optimizer.param_groups[0]['lr'] < params['min_lr']: print("\n!! LR SMALLER OR EQUAL TO MIN LR THRESHOLD.") break _, test_acc = evaluate_network(model, device, test_loader, epoch) _, train_acc = evaluate_network(model, device, train_loader, epoch) print("Test Accuracy: {:.4f}".format(test_acc)) print("Train Accuracy: {:.4f}".format(train_acc)) return train_acc, test_acc