COMBINING PHYSICS AND MACHINE LEARNING  FOR NETWORK FLOW ESTIMATION

This folder contains code, data and results associated to our ICLR'21 submission.

The folder structure is the following:
	- code: Contains .py files with python code and .ipynb files with jupyter
	notebooks used to run our experiments. Important code dependencies:
		- pytorch
		- higher: https://github.com/facebookresearch/higher/
		- DGL: https://github.com/dmlc/dgl
	Our models are implemented mostly in bilevel_mlp.py and bilevel_gcn.py

	- data: Contains the two datasets (Traffic and Power) used in our 
	experiments.

	- models: Contains the learned regularizers (MLPs) shown in the
	paper (as .pt files).

	- results: Contains all the plots in the paper.
