No need batch size parameter on simulator

furcelay / gigalens

Gradient Informed, GPU Accelerated Lens modelling (GIGALens) -- a package for fast Bayesian inference on strong gravitational lenses.

https://giga-lens.github.io/gigalens

MIT License

0 stars 0 forks source link

No need batch size parameter on simulator #1

Closed furcelay closed 1 month ago

furcelay commented 6 months ago

The Simulator class may have a rightmost batch dimension with length 1 that accepts any batch size. This may avoid saving large arrays as tree constants and making a reusable simulator. Test if it causes retracing.

furcelay commented 1 month ago

it is needed for EPL profile