search

ErikOsinga / master_research_project

Repository for the Master Research Project: "Data compression for weak lensing studies with the upcoming Euclid mission with Information Maximizing Neural Networks"
0 stars 0 forks source link

Fitting the summaries #9

Open mjvakili opened 5 years ago

mjvakili commented 5 years ago

Once, you have trained the network, can you run it on ~ 50 cosmologies that are distributed in the (omega_m , sigma_8) plane according to latin hypercube?

Then, you can look fit a parametric function to the outputs x1 & x2 as a function of omega_m , sigma_8.

ErikOsinga commented 5 years ago

The function that you suggested for constructing the latin hypercube does not seem to work..

http://www.columbia.edu/~ap3020/LensTools/html/examples/design.html

It seems that this "Design" object does not have the methods that are called in this example. I cannot find "add_parameter" or "put_points" anywhere.

ErikOsinga commented 5 years ago

The latin hypercube in the Omega_m, sigma_8 plane results in these output summaries.

LHS_5_1D_withnoise

These are quite hard to fit with a simple plane, so the posteriors I derived from doing that look horrible..

ErikOsinga commented 5 years ago

I think I got some very cool results that we will discuss tomorrow,

I fit a Gaussian Process to the 50 summaries generated from a Latin hyper-cube design, and used the Gaussian Process as the forward simulator (like an emulator) to do ABC. The blue posterior is inferred from this fit, and the orange posterior is inferred from a full 20,000 forward simulations. We can see that they agree very well! So it seems we do not need to do a lot of full forward simulations!

ABC_comparison