multiprobe-simulation-inference

Collection of inference methods to go from arbitrary summary statistics (neural network, peaks, power spectrum, ...) to posterior parameter constraints. Inference and neural density estimation methods include:

• Normalizing Flows: Conditional implementation from FlowConductor
• Gaussian Mixture Models: As a simpler baseline neural density estimator.
• Gaussian Process Approximate Bayesian Computation: As an alternative to standard SBI methods [Fluri et al. 2021]

Installation

Requires Python >= 3.8, PyTorch (for normalizing flows), and optionally TensorFlow >= 2.0/TensorFlow-Probability (for Gaussian mixture models).

Main dependencies:

• multiprobe-simulation-forward-model for utilities and data loading
• y3-deep-lss for neural network summary statistics preprocessing

Step 1: Install companion packages from GitHub

# Install multiprobe-simulation-forward-model
pip install git+https://github.com/des-science/multiprobe-simulation-forward-model.git

# Install y3-deep-lss
pip install git+https://github.com/des-science/y3-deep-lss.git

Step 2: Install this package

On HPC clusters with pre-installed PyTorch (recommended):

pip install -e .

On systems without PyTorch:

pip install -e .[torch]

To include TensorFlow for Gaussian mixture models:

pip install -e .[torch,tf]

Use the first option when PyTorch is available via system modules (e.g., module load pytorch) to preserve optimized GPU configurations.

Repository Structure

msi

• msi/apps - Inference scripts for normalizing flow training and MCMC sampling
• msi/flow_conductor - Normalizing flow implementation using PyTorch and enflows
• msi/gaussian_mixture - Gaussian mixture model implementation using TensorFlow Probability
• msi/utils - MCMC sampling, preprocessing, diagnostics, and visualization utilities
• msi/likelihood_base.py - Base class for likelihood implementations

configs

Configuration files for inference settings and hyperparameters.

data

Stored chains from DES Y3 analyses and figures.

notebooks

Notebooks for simulation-based inference via neural likelihood estimation and MCMC sampling.

Companion Repositories

• Forward modeling: multiprobe-simulation-forward-model
• Informative map-level neural summary statistics: y3-deep-lss