Partially Input Convex Neural Networks in Portfolio Optimization

Project Structure

├───configs # .yml setup file for anaconda environment 
├───src
│   └───core
        └───simulation_main.py  # Configure and run simulation parameters
        └───simulation_evaluation_main.ipynb  # View simulation results    
│       ├───layers # Contains PICNN & ICNN model architecture
│       ├───optimization # Contains algorithms Bundle Entropy, Projected Newton & PDIPM
│       ├───simulation
            └───simulation.py  # Deep-Q-Learning algorithm
            # Contains further simulation architecture components 
│       └───utils # Contains various helper functions
├───simulation_data # Data with the 3 examples for the jupyter notebooks

Simulation Setup

1. Clone repository

Follow https://help.github.com/en/github/creating-cloning-and-archiving-repositories/cloning-a-repository on how to clone the remote repository to local machine

2. Create Anaconda Environment

Follow https://docs.conda.io/projects/conda/en/latest/user-guide/tasks/manage-environments.html on how to create a anaconda environment from the provided .yml file in python 3.7

3. Adjust simulation parameters

In src/core/simulation_main.py adjust simulation parameters

Note: Change the log path to store the simulation data

4. Run sumulation

Save changes in simulation_main.py and run the script in conda environment

5. View Results

Load log data in simulation_evaluation_main.ipynb Jupyter Notebook and view the simulation data

Reference

B. Amos, L. Xu, and J. Z. Kolter, Input convex neural networks, in Proceedings of the 34th International Conference on Machine Learning-Volume 70, 2017

Paper: https://arxiv.org/pdf/1609.07152.pdf

Repository: https://github.com/locuslab/icnn