pybird.emulator module

class pybird.emulator.Emulator(emu_path, knots_path, load_models=True, co=<pybird.common.Common object>)[source]

Bases: object

A class to emulate EFT of LSS calculations using neural networks.

The Emulator class provides fast predictions for various components of the EFT calculations by using pre-trained neural network models. It handles the emulation of loop corrections, counterterms, and IR resummation components.

emu_path

Path to directory containing emulator model files.

Type:

str

knots_path

Path to file containing emulator knot points.

Type:

str

kmax_emu

Maximum k value for which the emulator is valid.

Type:

float

co

Common parameters shared across calculations.

Type:

Common

knots

Array of knot points used in the emulation.

Type:

ndarray

logknots

Log of the knot points.

Type:

ndarray

emu_ploopl_mono

Emulator model for monopole loop corrections.

emu_ploopl_quad

Emulator model for quadrupole loop corrections.

emu_ploopl_hex

Emulator model for hexadecapole loop corrections.

emu_IRPs11

Emulator model for IR resummation P11 terms.

emu_IRPsct

Emulator model for IR resummation counterterms.

emu_IRPsloop_mono

Emulator model for monopole IR resummation loop terms.

emu_IRPsloop_quad

Emulator model for quadrupole IR resummation loop terms.

emu_IRPsloop_hex

Emulator model for hexadecapole IR resummation loop terms.

load_models()[source]

Load all emulator models from the specified path.

make_params()[source]

Prepare input parameters for emulator predictions.

setPsCfl()[source]

Set power spectrum and correlation function using emulator predictions.

PsCf_resum()[source]

Apply IR resummation to emulated results.

Initialize the Emulator for fast EFT calculations.

Parameters:

  • emu_path (str) – Path to directory containing trained emulator model files

  • knots_path (str) – Path to file containing emulator knot points

  • load_models (bool, optional) – Whether to load models immediately upon initialization, by default True

  • co (Common, optional) – Common parameters object, by default co

Notes

The emulator provides 1000x speedup over full EFT calculations by using pre-trained neural networks. Models are valid up to kmax_emu = 0.4 h/Mpc.

load_models()[source]

Load all pre-trained emulator models from disk.

Notes

This method loads neural network models for: - Loop corrections (monopole, quadrupole, hexadecapole) - IR resummation terms (P11, counterterms, loop terms)

Models are stored as HDF5 files and compiled for JAX execution.

make_params(kk, pk, f=1.0, time=False, ir=False, pca=False)[source]

Prepare input parameters for emulator predictions.

Parameters:

  • kk (ndarray) – k-array in h/Mpc

  • pk (ndarray) – Linear power spectrum in (Mpc/h)^3

  • f (float, optional) – Growth rate for RSD effects, by default 1.0

  • time (bool, optional) – Whether to include time dependence, by default False

  • ir (bool, optional) – Whether this is for IR resummation, by default False

  • pca (bool, optional) – Whether to use PCA compression, by default False

Returns:

Formatted input parameters for the emulator models

Return type:

ndarray

setPsCfl(bird, kk, pk, time, make_params=True)[source]

Set power spectrum and correlation function using emulator predictions.

Uses neural network emulators to predict loop corrections and sets them in the provided Bird object.

Parameters:

  • bird (Bird) – Bird object to store the emulated results

  • kk (ndarray) – k-array in h/Mpc

  • pk (ndarray) – Linear power spectrum in (Mpc/h)^3

  • time (bool) – Whether to include time dependence

  • make_params (bool, optional) – Whether to prepare input parameters, by default True

Notes

This method emulates the computationally expensive loop calculations, providing predictions for P22, P13, and correlation function multipoles.

PsCf_resum(bird, kk, pk, f, time, make_params=True, ir=True, pca=False)[source]