The running effect of the model is not as good as that stated in the paper

[TigerCatRabbit]

I use the same hyperparameter as the paper claimed,but got low acc。The experimental results are lower than the original gcn。 2022-07-19 16:23:15,708 INFO Best validation acc: 0.79 Best test acc: 0.796 Final test acc: 0.796 2022-07-19 16:23:15,708 INFO Best model name: 123_vacc_0.79_tacc_0.796.pt Here is the hyperparameter : Namespace(alpha=0.15, attn_drop=0.5, clip=1.0, cuda=True, dataset='cora', do_train=True, early_stop=True, edge_drop=0.1, epochs=1500, fastmode=False, feed_forward=True, head_tail_shared=1, hop_num=8, in_drop=0.25, layer_norm=True, lr=0.00015, lr_reduce_factor=0.5, lr_schedule_patience=25, negative_slope=0.2, num_heads=8, num_hidden=512, num_layers=6, p_norm=0.0, patience=300, project_dim=-1, residual=True, save_path='../models/', seed=2020, self_loop=1, top_k=5, topk_type='local', weight_decay=1e-05)