Comparing NPE and NLE

[Justinezgh]

Results I got for SNPE-b (starting from power spectrum posterior) with smooth NF and mixture of 7 NFs

For the same simulation budget, results I got for SNLE (starting from power spectrum posterior) with affine NF and only one NF

next steps:

• check how NLE with affine NF behaves with 1k simulations when starting from prior and not power spectrum
• find the correct score weight for SNLE to check if adding gradient reduce the number of sim
• check what happens with simulation budget < 1k

script can be found here