Forgotten/mnn-H

Multiscale Neural Network (MNN) based on hierarchical matrices

This repo is the code for the paper: https://arxiv.org/abs/1807.01883 The code to generate data is written by MATLAB and the neural network is implemented by python based on keras on top of tensorflow

How to run

• generate data by matlab code
• run the code in the tensorflow environment

For NLSE, the main file of the matlab code is NLSEsample.m for 1d and NLSEsample2d.m for 2d.

The neural network use argparse to set the parameters. One can use

	python testHmatrix.py --help  

to print the usage of all the parameters and its default values. One example to run the code:

	python testHmatrix.py --epoch 200 --alpha 4 --output-suffix V1

The code for Kohn-Sham map is same as that for NLSE, thus we only provide the code for NLSE.