Available pre-trained predictors#
The picasso
package comes with several pre-trained models, allowing direct inference of gas thermodynamics from halo properties and potential.
They are all accessible in the picasso.predictors
module.
576 runs#
The six models described in Kéruzoré+24.
Underlying cosmology: Planck 2020 (\(\Omega_{\rm cdm} = 0.26067\), \(\omega_{\rm b} = 0.02242\), \(h = 0.6766\), \(\sigma_8 = 0.8102\), \(n_s = 0.9665\), \(w = −1\)), \(\Omega_k = 0\), \(\Sigma m_\nu = 0\)
\(L_{\rm box} = 576 h^{−1} \; {\rm Mpc}\), \(L_{\rm box} = 576 \; h^{−1}{\rm Mpc}\), \(2304^3\) particles (Gravity-only particle mass: \(m_m = 1.34 \times 10^9 \; h^{-1}M_\odot\)).
Trained on \(M_{500c} > 10^{13.5} \; h^{-1}M_\odot\)
Full input vector (see details in table): \(\log_{10} (M_{200c} / 10^{14} \; h^{-1}M_\odot)\), \(c_{200c}\), \(\Delta x / R_{200c}\), \(c_{\rm acc.}/c_{200c}\), \(c_{\rm peak}/c_{200c}\), \(e\), \(p\), \(a_{\rm \rm lmm}\), \(a_{25}\), \(a_{50}\), \(a_{75}\), \(\dot{M} / (M_\odot / {\rm GYr})\).
Model name |
In-code reference |
Description |
---|---|---|
baseline |
|
Uses full input vector, trained to reproduce non-radiative hydrodynamics profiles, with fixed \(c_\gamma = 0\) |
compact |
|
Uses full input vector minus mass assembly history, trained to reproduce non-radiative hydrodynamics profiles, with fixed \(c_\gamma = 0\) |
minimal |
|
Uses halo mass and concentration, trained to reproduce non-radiative hydrodynamics profiles, with fixed \(c_\gamma = 0\) |
subgrid |
|
Uses full input vector, trained to reproduce full-physics hydrodynamics profiles, with fixed \(c_\gamma = 0\) |
NR + \(\Gamma(r)\) |
|
Uses full input vector, trained to reproduce non-radiative hydrodynamics profiles, with variable \(c_\gamma\) |
SG + \(\Gamma(r)\) |
|
Uses full input vector, trained to reproduce full-physics hydrodynamics profiles, with variable \(c_\gamma\) |
See also:
picasso.predictors: From halo properties to gas properties, for the documentation of the prediction functions available in
picasso
;Using the picasso trained predictors, for code examples.