Open zengsihang opened 8 months ago
Thanks for getting in touch! Successfully training SurvLatent ODE can depend on the specific data used. A strategy that we found generally helpful is to set the dimension of latent states to be larger than the dimension of the input data (refer to this work https://arxiv.org/abs/1904.01681 for more details). If you could share more details about the MIMIC-III data you're using and any error messages you've received during training, that'd be helpful.
Hi,
Thanks for the great work. I wonder the specific hyperparameters for the MIMIC III experiments. The paper seems only have the searching range of hyperparameters and I found the default hyperparmeters in the notebook seems to fail on MIMIC III data.
Thanks!