Our current dataloaders only load summaries and parameters from existing storage. However, if we want to do active learning, we want to be able to sample from our data loaders at specific cosmologies dictated by our training procedure. Then, later, we can integrate these dataloaders with our forward modeling methods, e.g. BORG-PM, pmwd, and AFM methods.
The goal here is to build a dataloader which acts similarly to our existing dataloaders, except that it can also sample new data points from an external simulator.
A good place to start would be building a simulator-dataloader for our toy example in example_inference.py.
Our current dataloaders only load summaries and parameters from existing storage. However, if we want to do active learning, we want to be able to sample from our data loaders at specific cosmologies dictated by our training procedure. Then, later, we can integrate these dataloaders with our forward modeling methods, e.g. BORG-PM, pmwd, and AFM methods.
The goal here is to build a dataloader which acts similarly to our existing dataloaders, except that it can also sample new data points from an external simulator.
A good place to start would be building a simulator-dataloader for our toy example in example_inference.py.