Open Diagle opened 4 months ago
Hi, we have made 200 as our default parameter. You can try to further tune it to get a better result.
Thank,but I cannot reproduce the same results of the benchmark of scdeal with the provided parameter when training model without checkpoint
When we reproduce scDEAL, we find it hard to achieve the results in your paper, but we can slowly improve the performance of the model by adjusting the batch_size, may I ask if you neglected to give the hyperparameter batch_size.