Add conditional LogZ calculation

GFNOrg / torchgfn

A modular, easy to extend GFlowNet library

https://torchgfn.readthedocs.io/en/latest/

Other

238 stars 31 forks source link

Open josephdviviano opened 12 months ago

josephdviviano commented 12 months ago

When using a conditional gflownet, we need the LogZ estimate (in algorithms such as TB) to be conditional, i.e., predicted from the state.

This means the logz parameter must not be constrained to be a single parameter.

Also, for conditional generation, sampling should accept an instance and/or iterator optionally, to condition on.

josephdviviano commented 2 months ago