This project contains minimal implementations of RNN architectures trained with Backpropagation through time (BPTT) and Reservoir Computing (RC) for high-dimensional time-series forecasting. The following models are implemented:
- Long short-term memory (LSTM) trained with BPTT
- Unitary RNNs trained with BPTT
- Reservoir computers (RC) or Echo-state-networks (ESN)
- Multilayered perceptron (feedforward MLP) based on a windowing approach.
Moreover, spatial parallelization of the aforementioned models are implemented according to [1]. For implementation of the method to compute the Lyapunov spectrum, please refer to the repository RNN-Lyapunov-Spectrum.
The code requires python 3.7.3, tensorflow 1.11.0 Other required packages are: matplotlib, sklearn, psutil.
- python 3.7.3
- tensorflow 1.11.0
- matplotlib, sklearn, psutil
- mpi4py (parallel implementations)
The packages can be installed as follows: you can create a virtual environment in Python3 with:
python3 -m venv venv-RNN-RC-Chaos
Then activate the virtual environment:
source ./venv-RNN-RC-Chaos/bin/activate
Install a version of tensorflow (paper was compiled with version 1.11, which may no longer be available), here we also tested a more recent verion 1.14, (apart from warnings the code should run fine):
pip install tensorflow==1.14.0
Install the rest of the required packages with:
pip3 install matplotlib sklearn psutil mpi4py
The code is ready to run, you can test the following demo.
Parallelized networks that take advantage of the local interactions in the state space employ MPI communication. After installing an MPI library (open-mpi or mpich), the mpi4py library can be installed with:
pip3 install mpi4py
The code to get the exact environment (no mpi4py/parallel models yet) used in the paper is:
pip install virtualenv
virtualenv venv-RNN-RC-Chaos --python=python3.7.3
source ./venv-RNN-RC-Chaos/bin/activate
pip3 install -r requirements.txt
In macOs to install mpi4py:
source ./venv-RNN-RC-Chaos/bin/activate
pushd /tmp
rm -f tmp.c && touch tmp.c
xcrun -sdk macosx clang -arch x86_64 -c tmp.c
export OMPI_CC=xcrun
export MPICH_CC=xcrun
pip install --no-cache-dir mpi4py
The data to run a small demo are provided in the local ./Data folder
In order to run the demo in a local cluster, you can navigate to the Experiments folder, and select your desired application, e.g. Lorenz3D. There are scripts for each model. For example, you can ran a Reservoir Computer (also called Echo state network) with the following commands:
cd ./Experiments/Lorenz3D/Local
bash 01_ESN_auto.sh.sh
A statefull GRU or a parallel ESN can be run with:
bash 04_RNNStatefull_GRU.sh
bash 05_ESN_Parallel.sh
After running the command, you will see at the terminal output the training/testing progress. You can then navigate to the folder ./Results/Lorenz3D and check the different outputs of each model.
This code was developed in the CSE-lab. For questions or to get in contact please refer to [email protected].
This is joint work with:
- Jaideep Pathak (website, scholar)
- Brian R. Hunt (website, scholar)
- Themis Sapsis (website, scholar)
- Michelle Girvan (website, scholar)
- Edward Ott (website, scholar)
- Petros Koumoutsakos (website, scholar)
[1] P.R. Vlachas, J. Pathak, B.R. Hunt et al., Backpropagation algorithms and Reservoir Computing in Recurrent Neural Networks for the forecasting of complex spatiotemporal dynamics. Neural Networks, 2020 (doi: https://doi.org/10.1016/j.neunet.2020.02.016.)
[2] J. Pathak, B.R. Hunt, M. Girvan, Z. Lu, and E. Ott, Model-Free Prediction of Large Spatiotemporally Chaotic Systems from Data: A Reservoir Computing Approach. Physical Review Letters 120 (2), 024102, 2018
[3] P.R. Vlachas, W. Byeon, Z.Y. Wan, T.P. Sapsis, and P. Koumoutsakos Data-driven forecasting of high-dimensional chaotic systems with long short-term memory networks. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences 474 (2213), 2018