Problem during the adaptation to graph data for PCL.

Question

Problem during the adaptation to graph data for PCL.

ha-lins opened this issue 3 years ago · 8 comments

Hi Junnan,

Thanks for your great work! Now I'd like to conduct the PCL experiments on graph classification task (i.e., TUDataset), which includes about 1-5k graph instances with two-classes labels. I modified the code based on GraphCL. However, the Acc@Proto is 0.0 and the eval results are unchangeable approximately. I printed this line and the accp = 0 and the proto_out & proto_target is as follows:

My hyperparameters are as follow:

    parser.add_argument('--num-cluster', default=[200], type=int,
                        help='number of clusters')
    parser.add_argument('--bs', dest='bs', type=int, default=128,
                        help='batch_size')
    queue size; number of negative keys (default: 128)

Could you help me locate the reason for this problem? Thanks in advance!

Answer 1 · 2021-04-27T04:55:42.000Z

Hi, the first 20 epochs are warm-up epochs with only InfoNCE loss, that's probably why the acc@proto is 0.

Answer 2 · 2021-04-27T07:14:34.000Z

Thanks for your reply. I've revised the --warmup-epoch to 1 and is there any other reason? Are the values of proto_out & proto_target reasonable? Thanks!

Answer 3 · 2021-04-27T07:41:28.000Z

it seems that each sample has similar similarity with many prototypes? e.g., the first 3 items of 2nd row of proto_out has the same value -0.0294, maybe you want to check if the prototypes have collapsed?

Answer 4 · 2021-04-27T08:31:10.000Z

Yes, could you pls tell me how to check if the prototypes have collapsed? I also printed the pos_proto_id & neg_proto_id & all_proto_id. Are these values reasonable?

Answer 5 · 2021-04-27T08:33:21.000Z

it seems most of your embeddings have collapsed to a few prototypes? you can save the prototype vectors (maybe also embeddings) and check if they represent meaningful cluster centroids.

Answer 6 · 2021-04-27T10:01:05.000Z

Yes, could you pls explain what the index means here? I think it is from the ImageFolderInstance here and is the class label of the image. Is such understanding right? I'm not very familiar with computer vision. Thanks!

Answer 7 · 2021-04-27T11:58:47.000Z

The index can be considered as image_id

Answer 8 · 2021-04-28T09:05:16.000Z

Thanks for your reply.
(1) I found there are not such graph_id in graph datasets. Can I replace the index with range(batch_size)? I wonder if the image_id is different in various batches. Thanks!

(2) Are the eval_dataset the same as the train_dataset except for the different augmentation? This setting is a bit confusing and why they are not different images?