[Model] CoBFormer #233
Open
[Model] CoBFormer #233
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
added 2 commits
March 25, 2025 23:08
Author
|
cobformer的内容已更新 |
gyzhou2000
reviewed
Mar 27, 2025
| @@ -0,0 +1 @@ | |||
| python cobformer_trainer.py --dataset=Cora --learning_rate=0.01 --gcn_wd=1e-3 --weight_decay=5e-5 --gcn_type=1 --gcn_layers=2 --n_patch=112 --use_patch_attn --alpha=0.7 --tau=0.3 --gpu_id=0 | |||
Contributor
There was a problem hiding this comment.
参考其他模型的readme文件，写一下数据集描述，运行的命令，结果
Comment on lines
62
to
66
| def fix_seed(seed): | ||
| random.seed(seed) | ||
| np.random.seed(seed) | ||
| tlx.set_seed(seed) | ||
| tlx.set_device(device=f'gpu:{args.gpu_id}', id=args.gpu_id) |
Comment on lines
184
to
187
| parser.add_argument('--num_hidden', type=int, default=64, help='隐藏层维度') | ||
| parser.add_argument('--num_layers', type=int, default=1, help='层数') | ||
| parser.add_argument('--n_head', type=int, default=4, help='注意力头数') | ||
| parser.add_argument('--num_epochs', type=int, default=500, help='训练轮数') |
|
|
||
| return res_gnn, res_trans | ||
|
|
||
| def run(args, device, data, patch, split_idx, alpha, tau, postfix): |
Comment on lines
208
to
266
| path = osp.join(osp.expanduser('~'), 'datasets', args.dataset) | ||
| n_patch = args.n_patch | ||
| alpha = args.alpha | ||
| tau = args.tau | ||
| load_path = None | ||
| if args.dataset in ['ogbn-products']: | ||
| load_path = f'Data/partition/{args.dataset}_partition_{n_patch}.pt' | ||
|
|
||
| postfix = "test" | ||
| runs = 5 | ||
| print("n_patch: ", n_patch) | ||
|
|
||
| dataset = Planetoid(path, args.dataset) | ||
| data = dataset[0] | ||
|
|
||
| data.train_mask = tlx.convert_to_numpy(data.train_mask) | ||
| data.val_mask = tlx.convert_to_numpy(data.val_mask) | ||
| data.test_mask = tlx.convert_to_numpy(data.test_mask) | ||
| # 在各个mask后面pad一个维度，pad的值为0，mask是一维数组，用np.pad(mask, (0, 1), mode='constant') | ||
| data.train_mask = np.pad(data.train_mask, (0, 1), mode='constant') | ||
| data.val_mask = np.pad(data.val_mask, (0, 1), mode='constant') | ||
| data.test_mask = np.pad(data.test_mask, (0, 1), mode='constant') | ||
|
|
||
| split_idx = { | ||
| 'train': data.train_mask, | ||
| 'valid': data.val_mask, | ||
| 'test': data.test_mask | ||
| } | ||
|
|
||
|
|
||
| patch = partition_patch(data, n_patch, load_path) | ||
| batch_size = args.batch_size | ||
|
|
||
| results = [[], []] | ||
| for r in range(runs): | ||
| res_gnn, res_trans = run(args, device, data, patch, split_idx, alpha, tau, postfix) | ||
| results[0].append(res_gnn) | ||
| results[1].append(res_trans) | ||
|
|
||
| print(f"==== Final GNN====") | ||
| result = tlx.convert_to_tensor(results[0]) * 100. # 替换torch.tensor | ||
| print(result) | ||
| print(f"max: {tlx.ops.reduce_max(result, axis=0)}") # 使用 tlx.ops.reduce_max | ||
| print(f"min: {tlx.ops.reduce_min(result, axis=0)}") # 使用 tlx.ops.reduce_min | ||
| print(f"mean: {tlx.ops.reduce_mean(result, axis=0)}") # 使用 tlx.ops.reduce_mean | ||
| print(f"std: {tlx.ops.reduce_std(result, axis=0)}") # 使用 tlx.ops.reduce_std | ||
|
|
||
| print(f'GNN Micro: {tlx.ops.reduce_mean(result, axis=0)[1]:.2f} ± {tlx.ops.reduce_std(result, axis=0)[1]:.2f}') | ||
| print(f'GNN Macro: {tlx.ops.reduce_mean(result, axis=0)[3]:.2f} ± {tlx.ops.reduce_std(result, axis=0)[3]:.2f}') | ||
|
|
||
| print(f"==== Final Trans====") | ||
| result = tlx.convert_to_tensor(results[1]) * 100. | ||
| print(result) | ||
| print(f"max: {tlx.ops.reduce_max(result, axis=0)}") | ||
| print(f"min: {tlx.ops.reduce_min(result, axis=0)}") | ||
| print(f"mean: {tlx.ops.reduce_mean(result, axis=0)}") | ||
| print(f"std: {tlx.ops.reduce_std(result, axis=0)}") | ||
|
|
||
| print(f'Trans Micro: {tlx.ops.reduce_mean(result, axis=0)[1]:.2f} ± {tlx.ops.reduce_std(result, axis=0)[1]:.2f}') |
Comment on lines
29
to
32
| pred1, pred2 = model(data.x, patch, data.edge_index, edge_weight=edge_weight, num_nodes=data.num_nodes) | ||
| loss = model.loss(pred1, pred2, label, split_index['train']) | ||
| loss.backward() | ||
| optimizer.step() |
Contributor
There was a problem hiding this comment.
这些接口要改用tlx的train_one_step接口
| from gammagl.models.cobformer import CoBFormer | ||
|
|
||
| def eval_f1(pred, label, num_classes): | ||
| # 将tlx tensor转换为numpy数组 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
Checklist
Please feel free to remove inapplicable items for your PR.
or have been fixed to be compatible with this change
Changes