README

==Code for generating and using Lyman-alpha forest emulators==

This is the code repository for arXiv:1812.04654, Bird et al 2019.
It handles creating simulation suites with carefully chosen points, generating simulation parameter files,
generating flux power spectra using fake_spectra and making an interpolator from these flux power spectra.
It also has a python likelihood function which can be fed to cobaya for: testing the convergence of the emulator with mock data, or sampling against BOSS DR14/DR9 data.

To adapt this to your own survey you want to subclass the Emulator class in coarse_grid.py and specify your
own parameter set and limits. An example of how to do this can be found in the KnotEmulator class. You may also
want to add your cluster (if it is not XSEDE's Stampede) to clusters.py in SimulationRunner

Useful submodules:
lyaemu.coarse_grid: handles the emulator generation.
lyaemu.likelihood: handles the likelihood sampling using cobaya.
lyaemu.latin_hypercube: generates points on a unit cube in a maximin latin hypercube
lyaemu.gpemulator: fits the interpolation using GPy.
lyaemu.mean_flux: models for the redshift evolution of the forest mean flux.
lyaemu.bayesian_opt: perform Bayesian optimization (acquisition function) on an emulator.

Other files:
coarse_grid_plot.py: Makes a few useful plots which show the accuracy of the emulator,
as well as the emulator's own error estimate.
flux_power.py: generates the flux power spectra, using native bins as a fraction of the simulation box
linear_theory.py: linear biasing model for the flux power spectrum
linear_emulator.py: emulates the linear biasing model about. For testing purposes, only lightly maintained.
lyman_data.py: loads flux power spectrum data, BOSS DR14 by default (DR9 available).
matter_power.py: simple sub-emulator for reading and emulating the matter power spectrum
plot_latin_hypercube.py: Small script for visualising latin hypercubes
quadratic_emulator.py: Subclass of Emulator that varies one parameter at a time, for quadratic polynomial interpolation.
tempdens.py: Prints the IGM thermal parameters, T0 and gamma.

plots/ : Contains scripts for making the plots in the paper. Data not included.
plots/make_emulators.py gives an example of how to generate an emulator
lyaemu/tests: Some simple unit tests.

Dependencies:
- Python 3
- GPy
- numpy
- scipy
- h5py
- matplotlib for the plots
- cobaya for the sampling
- fake_spectra module for generating flux power spectra from a simulation snapshot.
- SimulationRunner (not on pip: at https://github.com/sbird/SimulationRunner) for setting up MP-Gadget simulations.