run_zinc.py

import os, sys
os.environ["CUDA_VISIBLE_DEVICES"] = "1"
import time
from shutil import copy, rmtree
import torch.nn
from dataset_zinc import ZINC
import argparse
from tqdm import tqdm
# from k_gnn import TwoMalkin, ConnectedThreeMalkin, TwoLocal, ThreeMalkin, ThreeLocal
from zinc_models import *
from utils import create_subgraphs
from utils_edge_efficient import create_subgraphs as create_subgraphs_eff
from utils_edge_I2 import create_subgraphs2
import numpy as np
import random
from torch_geometric.utils import degree



# General settings.
parser = argparse.ArgumentParser(description='I2GNN for ZINC graphs')
parser.add_argument('--target', default=0, type=int) # 0 for detection of tri-cycle, 3,4,...,8 for counting of cycles
parser.add_argument('--filter', action='store_true', default=False,
                    help='whether to filter graphs with less than 7 nodes')
parser.add_argument('--convert', type=str, default='post',
                    help='if "post", convert units after optimization; if "pre", \
                    convert units before optimization')

# Base GNN settings.
parser.add_argument('--model', type=str, default='I2GNN')#NestedGIN_eff', help = 'GNN/NestedGIN_eff/I2GNN')
parser.add_argument('--layers', type=int, default=6)

# Nested GNN settings
parser.add_argument('--h', type=int, default=3, help='hop of enclosing subgraph;\
                    if None, will not use NestedGNN')
parser.add_argument('--max_nodes_per_hop', type=int, default=None)
parser.add_argument('--node_label', type=str, default='spd',
                    help='apply distance encoding to nodes within each subgraph, use node\
                    labels as additional node features; support "hop", "drnl", "spd", \
                    for "spd", you can specify number of spd to keep by "spd3", "spd4", \
                    "spd5", etc. Default "spd"=="spd2".')
parser.add_argument('--use_rd', action='store_true', default=True,
                    help='use resistance distance as additional node labels')
parser.add_argument('--subgraph2_pooling', default='mean-center-side', help='support mean, add and center\
                    for some models, default mean for most models')
parser.add_argument('--subgraph_pooling', default='mean-context', help='support mean and add\
                    for some models, default mean for most models')
parser.add_argument('--use_pooling_nn', action='store_true', default=False,
                    help='use additional MLPs between each pooling layer')
parser.add_argument('--virtual_node', action='store_true', default=False)
parser.add_argument('--double_pooling', action='store_true', default=True)
parser.add_argument('--gate', action='store_true', default=True)

# Training settings.
parser.add_argument('--epochs', type=int, default=1000)
parser.add_argument('--batch_size', type=int, default=256)
parser.add_argument('--lr', type=float, default=1E-3)
parser.add_argument('--lr_decay_factor', type=float, default=0.95)
parser.add_argument('--patience', type=int, default=10)
parser.add_argument('--drop_ratio', type=float, default=0.0)
# Other settings.
parser.add_argument('--normalize_x', action='store_true', default=False,
                    help='if True, normalize non-binary node features')
parser.add_argument('--squared_dist', action='store_true', default=False,
                    help='use squared node distance')
parser.add_argument('--not_normalize_dist', action='store_true', default=False,
                    help='do not normalize node distance by max distance of a molecule')
parser.add_argument('--use_max_dist', action='store_true', default=False,
                    help='use maximum distance between all nodes as a global feature')
parser.add_argument('--use_pos', action='store_true', default=False,
                    help='use node position (3D) as continuous node features')
parser.add_argument('--RNI', action='store_true', default=False,
                    help='use node randomly initialized node features in [-1, 1]')
parser.add_argument('--use_relative_pos', action='store_true', default=False,
                    help='use relative node position (3D) as continuous edge features')
parser.add_argument('--self_loop', action='store_true', default=False)
parser.add_argument('--seed', type=int, default=1)
parser.add_argument('--save_appendix', default='',
                    help='what to append to save-names when saving results')
parser.add_argument('--keep_old', action='store_true', default=False,
                    help='if True, do not overwrite old .py files in the result folder')
parser.add_argument('--dataset', default='zinc')
parser.add_argument('--load_model', default=None)
parser.add_argument('--eval', default=0, type=int)
parser.add_argument('--train_only', default=0, type=int)
args = parser.parse_args()

# set random seed
from torch_geometric.seed import seed_everything
torch.manual_seed(args.seed)
if torch.cuda.is_available():
    torch.cuda.manual_seed(args.seed)
random.seed(args.seed)
np.random.seed(args.seed)
seed_everything(args.seed)
# define dataloader (different for 3-WL)
device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")



if args.save_appendix == '':
    args.save_appendix = '_' + time.strftime("%Y%m%d%H%M%S")
# args.res_dir = 'results/QM9_{}{}'.format(args.target, args.save_appendix)
args.res_dir = 'results/' + args.dataset + '_' + args.model + args.save_appendix
print('Results will be saved in ' + args.res_dir)
if not os.path.exists(args.res_dir):
    os.makedirs(args.res_dir)
# Backup python files.
copy('run_zinc.py', args.res_dir)
copy('utils.py', args.res_dir)
copy('zinc_models.py', args.res_dir)
# Save command line input.
cmd_input = 'python ' + ' '.join(sys.argv) + '\n'
with open(os.path.join(args.res_dir, 'cmd_input.txt'), 'a') as f:
    f.write(cmd_input)
print('Command line input: ' + cmd_input + ' is saved.')



path = 'data/ZINC'


subgraph_pretransform = None
pre_transform = None
if args.h is not None:
    if type(args.h) == int:
        path += '/ngnn_h' + str(args.h)
    elif type(args.h) == list:
        path += '/ngnn_h' + ''.join(str(h) for h in args.h)
    path += '_' + args.node_label
    if args.use_rd:
        path += '_rd'
    if args.max_nodes_per_hop is not None:
        path += '_mnph{}'.format(args.max_nodes_per_hop)
    def pre_transform(g):
        return create_subgraphs(g, args.h,
                                max_nodes_per_hop=args.max_nodes_per_hop,
                                node_label=args.node_label,
                                use_rd=args.use_rd,
                                subgraph_pretransform=subgraph_pretransform)
    def pre_transform2(g):
        return  create_subgraphs_eff(g, args.h,
                                 max_nodes_per_hop=args.max_nodes_per_hop,
                                 use_rd=args.use_rd,
                                 subgraph_pretransform=subgraph_pretransform,
                                 self_loop=args.self_loop)
    def pre_transform3(g):
        return create_subgraphs2(g, args.h,
                                 max_nodes_per_hop=args.max_nodes_per_hop,
                                 node_label=args.node_label,
                                 use_rd=args.use_rd,
                                 subgraph_pretransform=subgraph_pretransform,
                                 center_idx=True, self_loop=args.self_loop)
pre_filter = None
if args.model == 'GNN':
    processed_name = 'processed'
    my_pre_transform = None
    print('Loading from %s' % "processed")
elif args.model == 'NGNN':
    processed_name = 'processed_n_h'+str(args.h)+"_"+args.node_label
    my_pre_transform = pre_transform
elif args.model == 'NestedGIN_eff':
    processed_name = 'processed_eff_h'+str(args.h)+"_"+args.node_label
    if args.h is not None:
        my_pre_transform = pre_transform2
    else:
        my_pre_transform = None
elif args.model == 'I2GNN':
    processed_name = 'processed_I2_h'+str(args.h)+"_"+args.node_label
    my_pre_transform = pre_transform3
else:
    print('Error: no such model!')
    exit(1)

if args.use_rd:
    processed_name = processed_name + '_rd'
if args.self_loop:
    processed_name = processed_name + '_self_loop'

if args.model == 'I2GNN' or args.model == 'NGNN':
    from dataloader_I2 import DataLoader  # use a custom dataloader to handle subgraphs
else:
    from dataloader import DataLoader  # use a custom dataloader to handle subgraphs

data_path = './data/zinc' if args.dataset=='zinc' else './data/zinc-full'

t11 = time.time()
train_dataset = ZINC(root=data_path, processed_name=processed_name, pre_transform=my_pre_transform, dataset='train')
test_dataset = ZINC(root=data_path, processed_name=processed_name, pre_transform=my_pre_transform, dataset='test')
val_dataset = ZINC(root=data_path, processed_name=processed_name, pre_transform=my_pre_transform, dataset='val')
t12 = time.time()
print("Preprocessing time cost: {}s,".format(t12 - t11))

# for dataset in [train_dataset, test_dataset, val_dataset]:
    ## choice 1: drop second identifier
    #dataset.data.z[:, 2:] = torch.zeros_like(dataset.data.z[:, 2:])
    #dataset.data.rd[:, 1] = torch.zeros_like(dataset.data.rd[:, 0])
    #dataset.data.center_idx[:, -1] = dataset.data.center_idx[:, 0]

    ## choice 1: drop all identifiers
    # dataset.data.rd = torch.zeros_like(dataset.data.rd)
    # dataset.data.z = torch.zeros_like(dataset.data.z)



# Normalize targets to mean = 0 and std = 1. data leaking?
# tenpercent = int(len(dataset) * 0.1)

# train_dataset = dataset[int(2 * tenpercent):]
y_train_val = torch.cat([train_dataset.data.y, val_dataset.data.y], dim=0)
mean = y_train_val.mean(dim=0)
std = y_train_val.std(dim=0)
train_dataset.data.y = (train_dataset.data.y - mean) / std
val_dataset.data.y = (val_dataset.data.y - mean) / std
test_dataset.data.y = (test_dataset.data.y - mean) / std
print('Mean = %.3f, Std = %.3f' % (mean, std))

cont_feat_start_dim = 5
if args.normalize_x:
    x_mean = train_dataset.data.x[:, cont_feat_start_dim:].mean(dim=0)
    x_std = train_dataset.data.x[:, cont_feat_start_dim:].std(dim=0)
    x_norm = (train_dataset.data.x[:, cont_feat_start_dim:] - x_mean) / x_std
    dataset.data.x = torch.cat([dataset.data.x[:, :cont_feat_start_dim], x_norm], 1)

# Compute the in-degree histogram tensor
# Compute the maximum in-degree in the training data.
#max_degree = -1
#for data in train_dataset:
#    d = degree(data.edge_index[1], num_nodes=data.num_nodes, dtype=torch.long)
#    max_degree = max(max_degree, int(d.max()))

#deg = torch.zeros(max_degree + 1, dtype=torch.long)
#for data in train_dataset:
#    d = degree(data.edge_index[1], num_nodes=data.num_nodes, dtype=torch.long)
#    deg += torch.bincount(d, minlength=deg.numel())

test_loader = DataLoader(test_dataset, batch_size=args.batch_size)
val_loader = DataLoader(val_dataset, batch_size=args.batch_size)
train_loader = DataLoader(train_dataset, batch_size=args.batch_size, shuffle=True)

kwargs = {
    'num_layers': args.layers,
    'subgraph_pooling': args.subgraph_pooling,
    'subgraph2_pooling': args.subgraph2_pooling,
    'use_pos': args.use_pos,
    'edge_attr_dim': 5,
    'use_max_dist': args.use_max_dist,
    'use_rd': args.use_rd,
    'RNI': args.RNI,
    'drop_ratio': args.drop_ratio,
    'use_pooling_nn': args.use_pooling_nn,
    'use_virtual_node': args.virtual_node,
    'double_pooling': args.double_pooling,
    'gate': args.gate,
}

model = eval(args.model)(train_dataset, **kwargs)
if args.load_model != None:
    cpt = torch.load(args.load_model)
    model.load_state_dict(cpt)
print('Using ' + model.__class__.__name__ + ' model')
model = model.to(device)

optimizer = torch.optim.Adam(model.parameters(), lr=args.lr)
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(
    optimizer, mode='min',factor=args.lr_decay_factor, patience=args.patience, min_lr=0.00001)


def train(epoch):
    model.train()
    loss_all = 0

    for t, data in enumerate(train_loader):
        if type(data) == dict:
            data = {key: data_.to(device) for key, data_ in data.items()}
            num_graphs = data[args.h[0]].num_graphs
        else:
            data = data.to(device)
            num_graphs = data.num_graphs
        optimizer.zero_grad()
        y = data.y
        y = y.view([y.size(0), 1])
        # Loss = torch.nn.MSELoss()
        Loss = torch.nn.L1Loss()
        loss = Loss(model(data), y)
        loss.backward()
        loss_all += loss * num_graphs
        optimizer.step()
    return loss_all / len(train_loader.dataset)


def test(loader):
    model.eval()
    with torch.no_grad():
        model.eval()
        error = 0
        for data in loader:
            if type(data) == dict:
                data = {key: data_.to(device) for key, data_ in data.items()}
            else:
                data = data.to(device)
            y = data.y
            y_hat = model(data)[:,0]
            error += torch.sum(torch.abs(y_hat - y))
        num = len(loader.dataset)
    return error / num * std


def loop(start=1, best_val_error=None):
    t1 = time.time()
    pbar = tqdm(range(start, args.epochs+start))
    count = 0
    for epoch in pbar:
        pbar.set_description('Epoch: {:03d}'.format(epoch))
        lr = scheduler.optimizer.param_groups[0]['lr']
        loss = train(epoch)
        val_error = test(val_loader)
        scheduler.step(val_error)
        count += 1
        if best_val_error is None:
            best_val_error = val_error
        if val_error <= best_val_error or count == 10:
            count = 0
            test_error = test(test_loader)
            best_val_error = val_error
            log = (
                    'Epoch: {:03d}, LR: {:7f}, Loss: {:.7f}, Validation MAE: {:.7f}, ' +
                    'Test MAE: {:.7f}, Test MAE norm: {:.7f}'
            ).format(
                epoch, lr, loss, val_error,
                test_error,
                test_error / std.cuda(),
            )
            print('\n'+log+'\n')
            with open(os.path.join(args.res_dir, 'log.txt'), 'a') as f:
                f.write(log + '\n')
    model_name = os.path.join(args.res_dir, 'model_checkpoint{}.pth'.format(epoch))
    torch.save(model.state_dict(), model_name)
    start = epoch + 1
    t2 = time.time()
    print("Preprocessing time cost: {}s, Training time cost: {}s".format(t12 - t11, t2 -t1))
    return start, best_val_error, log


best_val_error = None
start = 1
if args.eval:
    test_error, tp, fp, tn, fn, auc = test(test_loader)
    log = (
            'Test ACC: {:.5f}, TP|FP|TN|FN: {:.4f}|{:.4f}|{:.4f}|{:.4f}'
    ).format(
        test_error, tp, fp, tn, fn
        # test_error / std[target].cuda(),
        # test_error / conversion[int(args.target)].cuda() if args.convert == 'post' else 0
    )
    print(log)
else:
    start, best_val_error, log = loop(start, best_val_error)
    print(cmd_input[:-1])
    print(log)

# uncomment the below to keep training even reaching epochs
''' 
while True:
    start, best_val_error, log = loop(start, best_val_error)
    print(cmd_input[:-1])
    print(log)
    pdb.set_trace()
'''