Project LUNA-PASSA

This is a clone of the Final Project of AM207 - Stochastic Methods of Data Aanalysis, Inference and Optimization. The original repository can be found here.

Overview

Final project for AM207 at Harvard. LUNA stands for "Learned Uncertainty-Aware"; and PASSA for "Paulina, Avriel, Sangyoon, Shih-Yi, and Arthur".

Team Members

• Avriel Epps-Darling
• Paulina Toro Isaza
• Sangyoon Park
• Shih-Yi Tseng
• Arthur Young

Description

The goal of this project is to reproduce a SOTA method in published ICML work, "Learned Uncertainty-Aware (LUNA) Bases for Bayesian Regression using Multi-Headed Auxiliary Networks". In this project, we implemented LUNA and reproduced key results in this paper, and extended the scope to invistigate how different factors impact the behavior of LUNA. A complete report can be found in this notebook.

Directory Structure

├── README.md          <- The top-level README for developers using this project.
│
├── pyproject.toml     <- The file that defines build system, project metadata, and other
│                         install requirements.
│
├── poetry.lock        <- The file that resolves and downloads dependencies in `pyproject.toml`.
│
├── data
│
├── src                <- Source code for use in this project.
│
├── notebooks          <- Jupyter notebooks.
│
├── reports            <- Generated analysis as HTML, PDF, LaTeX, etc.
│
├── references         <- Data dictionaries, manuals, and all other explanatory materials.

Dependency Management

The current project repo uses poetry to manage dependencies among different Python packages, which is essential to reproducibility. Hence, all users of and contributors to the current project repo are expected to use poetry to install and/or update packages. Following are some basic commands that can get you started:

Once you clone the current repo into your local machine, you can run:

$ poetry install

to install the right versions of packages for running scripts in the project repo.

To use the new Python configuration that has been installed, you need to run:

$ poetry shell

which will activate the virtual environment for the project repo.

You can simply type:

$ exit

to exit from the virtual environment and return to the global (or system) Python installation.

If a new script that you are planning to add and commit to the repo requires a new package, you can run:

$ poetry add [package-name]

This will add the new package to the repo's virtual environment and update the corresponding information in the pyproject.toml and poetry.lock, which you need to add and commit as well.

Conversely, if changes to the project repo make a certain package not needed anymore, you can remove it by running:

$ poetry remove [package-name]

Once you set up the virtual environment using poetry, you can create the corresponding jupyter kernel as follows:

poetry run ipython kernel install --user --name=luna-passa

Running a notebook on this new kernel (luna-passa) will enable you to use the project-specific packages installed in the virtual environment.

You can check out the following resources to learn more about how to set up and use poetry:

• Official Documentation
• This tutorial walks you through how to install and use poetry along with pyenv (pyenv helps to better manage multiple versions of Python installation in the local machine).