You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Everything remains unchanged except for the batch number.
Is there something wrong with the settings or configuration?
The text was updated successfully, but these errors were encountered:
yyygou
changed the title
training is not converged on syn-balanced and enzymes datasets
Training is not converged on syn-balanced and enzymes datasets
May 31, 2024
Hello, I'm having some environment configuration issues and cannot run the training codes. Could you please share your python version and your package version? Thx a lot!
Hello, I'm having some environment configuration issues and cannot run the training codes. Could you please share your python version and your package version? Thx a lot!
hi, I am using Python 3.6 in windows platform, and installed the packages in the following order:
numpy-1.17.1
matplotlib
networkx
3.1 deepsnap-0.1.2
torch
scipy-1.0.1
pandas
seaborn
scikit_learn
torch-cluster-1.5.3
torch_scatter-2.0.3
torch-sparse-0.6.0
torch-spline-conv-1.2.0
pytest-runner-5.3.2
numba-0.53.1
torch-geometric-1.4.3
test_tube == 0.7.5
tqdm-4.43.0
if the version number is missing, pls use the suggested version number given by the requirements.txt
I also encountered the same problem, but some times the ACC will stuck at some other number, like 0.33 or 0.667 @yyygou May I ask have you resolved this issue?
Hi, it appears that I am encountering an issue while training a model on the syn-balanced and enzymes datasets.
In both cases, the training process seems to be running indefinitely, and the outputs are as follows:
Validation. Epoch 87. Acc: 0.5000. P: 0.5000. R: 1.0000. AUROC: 0.5000. AP: 0.5000.
TN: 0. FP: 2048. FN: 0. TP: 2048
Saving ckpt/model.pt
600it [00:00, 958.17it/s]. Training acc: 0.5000 000
600it [00:00, 980.59it/s] Batch 88394. Loss: 3.2000. Training acc: 0.5000
Batch 88999. Loss: 3.2000. Training acc: 0.5000
Everything remains unchanged except for the batch number.
Is there something wrong with the settings or configuration?
The text was updated successfully, but these errors were encountered: