This repository provides a fully functional pipeline for predicting the non-linear matter-power spectrum within parametrized gravity model. The pipeline leverages the COLA (COmoving Lagrangian Acceleration) method for simulations and applies machine learning tools to train an emulator using simulation data.
- Simulation code using the COLASolver from the FML library.
- Neural network training performed with PyTorch Lightning, a lightweight wrapper for the PyTorch module.
The repository is organized into the following folders:
Contains all scripts and configurations required to run the COLA simulations using the COLASolver implemented for including the parametrized gravity model.
Includes the machine learning component of the pipeline with scripts for training the neural network using PyTorch Lightning. This folder contains all the necessary code to:
- Draw parameter samples using Latin hypercube sampling.
- Create runfiles for the COLASolver code for all the samples.
- Convert COLASolver output data into a format suitable for neural network training.
- Prepare the input files for neural network training.
- Execute the neural network training process.