jean-eric

One way to make the maxK search for untruncated Moffat, is to use the approx 2D functions of (`beta`, `gsparam.maxk_threshold`) parameters ```python def nlm3(x,y): return -0.871303 \ + 2125.34*jnp.exp(548.446*x*y) \...

Hi Ismael It is not my final guess but I  think it is related to https://github.com/jecampagne/JaxTutos/blob/main/JAX_PyTree_initialisation.ipynb (section "A more complete ex with different kinds of variables") the point is when...

Hi! Erwan Allys point me your work, and I agree with @edouardoyallon and would support to use kymatio. Notice that the JAX backend would be a very nice by-product, indeed....

pyknos 0.15.1 Looking at the issue you pointed, I mention that I have dim=21 and use 100,000 simul per rounds. The crash occurs only during the 6th round but the...

Just a precision, I was running on jupyter nb. on Colab CPU

Ho Thanks a lot for your interest @michaeldeistler. I will test the multi-rounds with the single observation of the single optimisation round. In fact as was looking in the code...

Thanks a lot. Here is an update of my exercise: If I run 5 rounds of SNRE with 200 simul each ```python # 5 rounds: first round simulates from the...

Ha cool! One more question in the multi-rounds trip ```python theta, x = simulate_for_sbi(simulator, proposal, num_simulations=200) ``` is there a way to use your vectorized version `method="slice_np_vectorized", num_chains=100`?

heuu, I missed something ? I think the first proposal is the prior that do not have the acceelerator mechanism... isn't it? ```python 11 for _ in range(num_rounds): ---> 12...

Yes I 've update my script as I shown just before you :)