Code for our SIGCOMM'21 paper "Network Planning with Deep Reinforcement Learning".

Overview

0. Introduction

This repository contains the source code for our SIGCOMM'21 paper "Network Planning with Deep Reinforcement Learning".

Notes

The network topologies and the trained models used in the paper are not open-sourced. One can create synthetic topologies according to the problem formulation in the paper or modify the code for their own use case.

1. Environment config

AWS instance configurations

  • AMI image: "Deep Learning AMI (Ubuntu 16.04) Version 43.0 - ami-0774e48892bd5f116"
  • for First-stage: g4dn.4xlarge; Threads 16 in gurobi.env
  • for others (ILP, ILP-heur, Second-stage): m5zn.12xlarge; Threads 8 in gurobi.env

Step 0: download the git repo

Step 1: install Linux dependencies

sudo apt-get update
sudo apt-get install build-essential libopenmpi-dev libboost-all-dev

Step 2: install Gurobi

cd 
   
    /
./gurobi.sh
source ~/.bashrc

   

Step 3: setup && start conda environment with python3.7.7

If you use the AWS Deep Learning AMI, conda is preinstalled.

conda create --name 
   
     python=3.7.7
conda activate 
    

    
   

Step 4: install python dependencies in the conda env

cd 
   
    /spinninup
pip install -e .
pip install networkx pulp pybind11 xlrd==1.2.0

   

Step 5: compile C++ program with pybind11

cd 
   
    /source/c_solver
./compile.sh

   

2. Content

  • source
    • c_solver: C++ implementation with Gurobi APIs for ILP solver and network plan evaluator
    • planning: ILP and ILP-heur implementation
    • results: store the provided trained models and solutions, and the training log
    • rl: the implementations of Critic-Actor, RL environment and RL solver
    • simulate: python classes of flow, spof, and traffic matrix
    • topology: python classes of network topology (both optical layer and IP layer)
    • test.py: the main script used to reproduce results
  • spinningup
  • gurobi.sh
    • used to install Gurobi solver

3. Reproduce results (for SIGCOMM'21 artifact evaluation)

Notes

  • Some data points are time-consuming to get (i.e., First-stage for A-0, A-0.25, A-0.5, A-0.75 in Figure 8 and B, C, D, E in Figure 9). We provide pretrained models in /source/results/trained/ / , which will be loaded by default.
  • We recommend distributing different data points and differetnt experiments on multiple AWS instances to run simultaneously.
  • The default epoch_num for Figure 10, 11 and 12 is set to be 1024, to guarantee the convergence. The training process can be terminated manually if convergence is observed.

How to reproduce

  • cd /source
  • Figure 7: python test.py fig_7 , epoch_num can be set smaller than 10 (e.g. 2) to get results faster.
  • Figure 8: python test.py single_dp_fig8 produces one data point at a time (the default adjust_factor is 1).
    • For example, python test.py single_dp_fig8 ILP 0.0 runs ILP algorithm for A-0.
    • Pretrained models will be loaded by default if provided in source/results/trained/. To train from scratch which is NOT RECOMMENDED, run python test.py single_dp_fig8 False
  • Figure 9&13: python test.py single_dp_fig9 produces one data point at a time.
    • For example, python test.py single_dp_fig9 E NeuroPlan runs NeuroPlan (First-stage) for topology E with the pretrained model. To train from scratch which is NOT RECOMMENDED, run python test.py single_dp_fig9 E NeuroPlan False.
    • python test.py second_stage can load the solution from the first stage in and run second-stage with relax_factor= on topo . For example, python test.py second_stage D "results/ /opt_topo/***.txt" 1.5
    • we also provide our results of First-stage in results/trained/ / .txt , which can be used to run second-stage directly. For example, python test.py second_stage C "results/trained/C/C.txt" 1.5
  • Figure 10: python test.py fig_10 .
    • adjust_factor={0.0, 0.5, 1.0}, num_gnn_layer={0, 2, 4}
    • For example, python test.py fig_10 0.5 2 runs NeuroPlan with 2-layer GNNs for topology A-0.5
  • Figure 11: python test.py fig_11 .
    • adjust_factor={0.0, 0.5, 1.0}, mlp_hidden_size={64, 256, 512}
    • For example, python test.py fig_11 0.0 512 runs NeuroPlan with hidden_size=512 for topology A-0
  • Figure 12: python test.py fig_12 .
    • adjust_factor={0.0, 0.5, 1.0}, max_unit_per_step={1, 4, 16}
    • For example, python test.py fig_11 1.0 4 runs NeuroPlan with max_unit_per_step=4 for topology A-1

4. Contact

For any question, please contact hzhu at jhu dot edu.

Owner
NetX Group
Computer Systems Research Group at PKU
NetX Group
Implementation of paper "Decision-based Black-box Attack Against Vision Transformers via Patch-wise Adversarial Removal"

Patch-wise Adversarial Removal Implementation of paper "Decision-based Black-box Attack Against Vision Transformers via Patch-wise Adversarial Removal

4 Oct 12, 2022
A multilingual version of MS MARCO passage ranking dataset

mMARCO A multilingual version of MS MARCO passage ranking dataset This repository presents a neural machine translation-based method for translating t

75 Dec 27, 2022
Code for NeurIPS 2020 article "Contrastive learning of global and local features for medical image segmentation with limited annotations"

Contrastive learning of global and local features for medical image segmentation with limited annotations The code is for the article "Contrastive lea

Krishna Chaitanya 152 Dec 22, 2022
KSAI Lite is a deep learning inference framework of kingsoft, based on tensorflow lite

KSAI Lite is a deep learning inference framework of kingsoft, based on tensorflow lite

80 Dec 27, 2022
The official PyTorch code implementation of "Personalized Trajectory Prediction via Distribution Discrimination" in ICCV 2021.

Personalized Trajectory Prediction via Distribution Discrimination (DisDis) The official PyTorch code implementation of "Personalized Trajectory Predi

25 Dec 20, 2022
Fast, modular reference implementation and easy training of Semantic Segmentation algorithms in PyTorch.

TorchSeg This project aims at providing a fast, modular reference implementation for semantic segmentation models using PyTorch. Highlights Modular De

ycszen 1.4k Jan 02, 2023
ANEA: Automated (Named) Entity Annotation for German Domain-Specific Texts

ANEA The goal of Automatic (Named) Entity Annotation is to create a small annotated dataset for NER extracted from German domain-specific texts. Insta

Anastasia Zhukova 2 Oct 07, 2022
CVPR 2021 Challenge on Super-Resolution Space

Learning the Super-Resolution Space Challenge NTIRE 2021 at CVPR Learning the Super-Resolution Space challenge is held as a part of the 6th edition of

andreas 104 Oct 26, 2022
Pytorch code for "Text-Independent Speaker Verification Using 3D Convolutional Neural Networks".

:speaker: Deep Learning & 3D Convolutional Neural Networks for Speaker Verification

Amirsina Torfi 114 Dec 18, 2022
Liquid Warping GAN with Attention: A Unified Framework for Human Image Synthesis

Liquid Warping GAN with Attention: A Unified Framework for Human Image Synthesis, including human motion imitation, appearance transfer, and novel view synthesis. Currently the paper is under review

2.3k Jan 05, 2023
Towards Representation Learning for Atmospheric Dynamics (AtmoDist)

Towards Representation Learning for Atmospheric Dynamics (AtmoDist) The prediction of future climate scenarios under anthropogenic forcing is critical

Sebastian Hoffmann 4 Dec 15, 2022
Parallel Latent Tree-Induction for Faster Sequence Encoding

FastTrees This repository contains the experimental code supporting the FastTrees paper by Bill Pung. Software Requirements Python 3.6, NLTK and PyTor

Bill Pung 4 Mar 29, 2022
thundernet ncnn

MMDetection_Lite 基于mmdetection 实现一些轻量级检测模型,安装方式和mmdeteciton相同 voc0712 voc 0712训练 voc2007测试 coco预训练 thundernet_voc_shufflenetv2_1.5 input shape mAP 320

DayBreak 39 Dec 05, 2022
Nvidia Semantic Segmentation monorepo

Paper | YouTube | Cityscapes Score Pytorch implementation of our paper Hierarchical Multi-Scale Attention for Semantic Segmentation. Please refer to t

NVIDIA Corporation 1.6k Jan 04, 2023
Portfolio Optimization and Quantitative Strategic Asset Allocation in Python

Riskfolio-Lib Quantitative Strategic Asset Allocation, Easy for Everyone. Description Riskfolio-Lib is a library for making quantitative strategic ass

Riskfolio 1.7k Jan 07, 2023
Automatic deep learning for image classification.

AutoDL AutoDL automates machine learning tasks enabling you to easily achieve strong predictive performance in your applications. With just a few line

wenqi 2 Oct 12, 2022
Official implementation of Influence-balanced Loss for Imbalanced Visual Classification in PyTorch.

Official implementation of Influence-balanced Loss for Imbalanced Visual Classification in PyTorch.

Seulki Park 70 Jan 03, 2023
PyTorch Implementation of Backbone of PicoDet

PicoDet-Backbone PyTorch Implementation of Backbone of PicoDet Original Implementation is implemented on PaddlePaddle. Example picodet_l_backbone = ES

Yonghye Kwon 7 Jul 12, 2022
TransZero++: Cross Attribute-guided Transformer for Zero-Shot Learning

TransZero++ This repository contains the testing code for the paper "TransZero++: Cross Attribute-guided Transformer for Zero-Shot Learning" submitted

Shiming Chen 6 Aug 16, 2022
DenseCLIP: Language-Guided Dense Prediction with Context-Aware Prompting

DenseCLIP: Language-Guided Dense Prediction with Context-Aware Prompting Created by Yongming Rao*, Wenliang Zhao*, Guangyi Chen, Yansong Tang, Zheng Z

Yongming Rao 321 Dec 27, 2022