Recent tests have indicated that the CDM field alone gives really poor parameter estimation, far poorer than the full results shown in the CAMELs papers (order 15% and 4% on Omega_m and sigma_8 respectively), and basically no constraining power on the feedback parameters. Given this we should look at expanding to include more/all fields (will see what I can get away with in terms of memory). This works already for the CAMELs CNN models, but not for the scattering modules.
Chris-Pedersen
commented
2 years ago
Recent tests have indicated that the CDM field alone gives really poor parameter estimation, far poorer than the full results shown in the CAMELs papers (order 15% and 4% on Omega_m and sigma_8 respectively), and basically no constraining power on the feedback parameters. Given this we should look at expanding to include more/all fields (will see what I can get away with in terms of memory). This works already for the CAMELs CNN models, but not for the scattering modules.