Online Learning Émulator

Documentation:

The Online Learning Émulator - OLÉ is a tool to efficiently accelerate statistical analyses with a focus on Cosmology.

It follows combines the idea of emulating computationally expensive codes with the idea of online learning. This is particularly useful when:

• The theory code is computationally expensive. In particular, if it is more expensive than the likelihood code
• A explicit likelihood is available
• The dimensionality of the theory code input does not exceed about 20

Features of OLÉ:

• Easy to use framework
• JAX based implementation
• Implementation of various sampling algorithms like Ensemble, Minimizer, NUTS
• Interfaces with inference tools like Cobaya and MontePython.

Installation

To install OLÉ run:

git clone git@github.com:svenguenther/OLE.git
cd OLE
pip install .

While not a strict requirement, mpi4py is recommended for running multiple parallel chains. You can either install this manually, or by installing OLÉ with:

pip install .[MPI]

If you plan to make modifications to OLÉ, it is recommended to install OLÉ in editable mode by including the -e flag when pip installing.

If you plan to use OLE with MontePython, you should change the contents of the file MP_PATH in OLE/interfaces/ to redirect to your MontePython directory.

Documentation

The documentation is available at ReadTheDocs.

Examples

Examples on the different features can be found in the example directory.