nxhoang562 / customized-fedml Goto Github PK

View Code? Open in Web Editor NEW

This is a fork of FedML framework with additional models

Shell 0.31% Python 83.91% CSS 0.12% HTML 0.07% Jupyter Notebook 15.59%

customized-fedml's Introduction

customized-FedML

This is a fork of FedML framework with additional models

How to use LockEdge and VAE-LSTM in this customized FedML framework

0. Prerequisites

Install and Configure EMQX broker

from repo emqx

If pre-built binary packages aren't available for your OS, build from source is recommended (requires erlang and some other packages)

Configuration:

In emqx.conf:
- Set mqtt.max_packet_size = 20MB
- Set mqtt.retain_available = true
In plugins/emqx_retainer.conf:
- Set retainer.max_payload_size = 20MB

Install required python packages

Install in anaconda on server (recommended) and Raspberry Pi's machine env or virtual env (no conda due to unavailable tensorflow package on armv7l repo)

1. Configuration

Clients and Server

In FedML/fedml_iot/cfg.py:
- Modify HOST variable to your MQTT broker's IP address in your network
- Modify APP_HOST variable to your server's IP address (Weight Aggregator)
Make sure all clients and server are on the same network

Clients

In run.sh:
- Modify server_ip to server's IP address,
- Modify client_uuid from 0 for each client
- Change run script for your model (VAE_LSTM_fedavg_rpi_client.py or lockedge_fedavg_rpi_client.py)

Server

Modify model configuration json file (for VAE-LSTM) to your needs

2. Run

Instructions are written in chronological order of executions

On Server

Run EMQ X broker:
- cd to bin folder (if built from source)
- ./emqx start (built from source) or emqx start (binary package)
Run Aggregator Server:
- Activate your installed environment
- VAE-LSTM Model:
  python VAE-LSTM-app.py --config $(dir to model config file) --num-client $(number of workers)
- LockEdge Model:
  python LCHA-app.py --client-num-per-round $(number of workers) --comm-round $(number of comm rounds)