BBB-FLIDS

This repository contains BBB-FLIDS, Basic Blockchain-Based Federated Learning Intrusion Detection System.

For detailed information about this project, please see the report.

Setup

You may want to see "Dummy Blockchain Platform" heading before following all instructions here.

Virtual Env

Although it's not required, it's recommended to create a Python virtualenv for blockchain libraries.

First install virtualenv:

$ pip install virtualenv

Note that you might need elevated privileges:

$ sudo pip install virtualenv

After that, create the virtual environment in .venv folder:

# Create a virtual environment:
$ virtualenv -p python3 .venv
# Alternatively, enable the virtual environment to use system-wide packages:
# This allows you to avoid re-installing large libraries, e.g., pytorch
$ virtualenv -p python3 --system-site-packages .venv

It's recommended that you go with the latter option if you already have pytorch installed system-wide.

Activate your new virtual environment:

$ source .venv/bin/activate

Note that you will need to run this command each time you restart the terminal.

Install Dependencies

Install the required dependencies from requirements.txt:

$ pip install -r requirements.txt

After that, you need to run install_solc.py file to install the Solidity compiler:

$ python install_solc.py

Dummy Blockchain Platform

Thanks to DummyPlatform module, you can run this project without blockchain. This is useful if:

• You don't want to install blockchain dependencies.
• Or you want to run the federated learning part only for better run-time performance.

Instead of installing all dependencies, install numpy, pytorch, sklearn, pandas, and tqdm (optional) only. These are included in fl-requirements.txt:

$ pip install -r fl-requirements.txt

In config.ini, make sure that platform is set to "dummy".

Now you should be able to run the project without blockchain.

Running

This will run the simulation with the settings given in config.ini:

$ python main.py

You can supply your own ini file as an argument:

$ python main.py custom.ini

Configuring

Most settings are set in .ini file. Please see the comments in the given config.ini file for details.

In order to configure the neural network:

1. Create a new class in ModelConfig.py. Copy one of the examples for quickstart.
2. In .ini file, set the model name field to the class name of your new model.

Plotting

If evaluate per epoch is enabled in configuration file, the model will be evaluated on the validation set after each global epoch (round). Using the following command, you can plot the previous results you got:

$ python main.py

For this, matplotlib needs to be installed, which is included in neither requirements.txt nor fl-requirements.txt. Thus, you need to install it manually:

$ pip install matplotlib

Test Scripts

This will test some functionality in Solidity contract:

$ python TestContract.py

StressTestContract.py is intended to measure how much gas is required for a given byte-size of the machine-learning model.

$ python StressTestContract.py