Thank you for releasing the code. I am using a custom dataset with 10k graphs. I tried to update the code to include the kl divergence during training to check if there is overfitting on the smaller dataset. While the PosMSE seems fine, the results for E_kl and X_kl always give a nan for the training samples. Can you please tell me if there is something wrong with my approach?
Thank you for releasing the code. I am using a custom dataset with 10k graphs. I tried to update the code to include the kl divergence during training to check if there is overfitting on the smaller dataset. While the PosMSE seems fine, the results for E_kl and X_kl always give a nan for the training samples. Can you please tell me if there is something wrong with my approach?
self.train_metrics = torchmetrics.MetricCollection([custom_metrics.PosMSE(), custom_metrics.XKl(), custom_metrics.EKl()])
In my training_step, I invoke
nll, log_dict = self.compute_train_nll_loss(pred, z_t, clean_data=dense_data)
Finally, the method definition is
Any help would be highly appreciated.
Best, Chinmay