Open BidhanChandra opened 5 months ago
Hi, sure I can provide that. Are you mainly interested in the base-model training or the uncertainty estimator? Easiest / fastest is to just use a fixed pre-trained model and plugin the GP on top.
I have generated geometries which are going out of distribution from the initial starting points. So, I need to train the model on initial geometries and then calculate the uncertainty for later
Hello, I would also find very useful to have a simple example on how to apply the GP over a pretrained model. In my case, I'm using nequip over an md17 npz dataset.
Hi, I would greatly appreciate it if you could provide some additional detailed information, such as the conda environment configurations and the YAML config file found at https://github.com/wollschl/uncertainty_for_molecules/blob/196009587f9d648745c5c2559c544d29b9e8ee6f/ue4mol/train/train_debug.py#L6 Without these information, it is difficult to run the code and evaluate this method.
Hi, I am really interested in calculating the uncertainty using graph neural networks and I am going through your paper. I was wondering if you can provide some ipynb notebook to train such models and get the uncertainty estimates. That will be really helpful