alekscipri / cosmo1gal

Data and scripts used for the cosmology with one galaxy paper

Prerequisites

The packages needed to run the codes are:

• python3
• numpy
• Pytorch
• Optuna
• XGB
• SHAP (only to compute SHAP values)
• Pylians3 (only for Fisher analysis)

If the user wants to add more galaxy properties to the ones considered here he/she will need to download the CAMELS Subfind catalogues.

Structure

There are four different folders:

• data: this folder contains the codes to generate the data and the data itself.
• neural_networks: this folder contains the codes and results obtained from analyzing the data with neural networks.
• XBG: this folder contains the codes and results obtained by analyzing the data with gradient boosting trees.
• other: this folder contains other codes written for the data analysis.

We now describe in a bit more detail the content of each folder.

data

This folder contains the following files:

• preprocess.py. This script will read the Subfind catalogues and generate the input files for both the neural networks and the gradient boosting trees. The output of this script are two files:
  • galaxies_X_z=Z.ZZ.txt. These files contain the galaxy properties for each galaxy in all simulations. X can be IllustrisTNG or SIMBA, and Z.ZZ is the redshift.
  • offset_X_z=Z.ZZ.txt. These files contain the offset to identify galaxies belonging to different simulations. This file is used to create the training, validation, and testing sets splitting galaxies across simulations. X can be IllustrisTNG or SIMBA, and Z.ZZ is the redshift.
• Omega_b.py. This script is similar to preprocess.py and is used to generate the galaxy and offset files for the simulations with different values of Omega_b. This script will generate the file:
  • galaxies_Omega_b.txt that contains the galaxy properties for the simulations with different values of Omega_b.
• latin_hypercube_params_X.txt. This file contains the value of the cosmological and astrophysical parameters for each simulation. X can be IllustrisTNG or SIMBA.

We note that the galaxies_* files are too heavy to be stored in GitHub. We provide access to them through:

• Globus
• Url

neural networks

This folder contains the codes, databases, and weights of the neural networks. There are different files:

• architecture.py. This script contains the different neural network architectures.
• data.py. This script reads the data and prepare it to train the networks.
• main_LFI.py. This script is used to train the networks.
• test_LFI.py. This script is used to test the networks.
• train_with_feedback_LFI.py. This script is used to train models where the value of the astrophysical parameters are known.
• analyze_databases.py. This script will read the different databases and print some information about their best trials.
• analyze_results.py. This script is used to analyze the results after training the networks.
• shap_values.py. This script is used to compute the shape values.

There are also different folders:

• databases. This folder contains the databases.
• losses. This folder contains the losses of the different models.
• models. This folder contains the network weights for the different models.
• Results.txt. This folder contains the results of testing the models.
• shap. This folder contains the SHAP values of the model.

Unfortunately, the folders are too heavy to be hosted in GitHub. We however provide access to them through:

• Globus
• Url

XGB

This folder contains the scripts, databases, and results of performing the analysis using gradient boosting trees.

other

This folder contains the codes used to carry out the Fisher matrix calculation.