Enable batches in profile generation

DifferentiableUniverseInitiative / GalFlow

A TensorFlow reimplementation of GalSim

MIT License

10 stars 4 forks source link

Enable batches in profile generation #15

Closed b-remy closed 2 years ago

b-remy commented 3 years ago

With this PR, we are now able to generate Gaussian and Sersic profiles by batch. The only requirement is to provide parameters by list of floats, tensors or numpy arrays (which are then converted in TF tensors).

An example can be seen at https://github.com/b-remy/gems/blob/main/toymodel1.py (752107e):

Figure_2

b-remy commented 3 years ago

Hi @EiffL , I just made a few more changes in order to:

enable tf.function
fix all the NaNs appearing for some Sersic profiles

EiffL commented 3 years ago

It works much better :-D but I haven't tried to run a chain with free hlr and n yet

EiffL commented 3 years ago

hummm still running into some weird problems when trying to run a chain over size and ellipticity

b-remy commented 3 years ago

It should be fixed now. I ran a chain with free n and hlr and no error occurred.