Instructions for MLDL23-FL

This is a set of instructions for getting started with the project.

Clone

First you need to clone the project into whatever directory you prefer.

git clone https://github.com/RicoBorra/FederatedLearningProject

Then, for the following operations, move into the project folder.

cd FederatedLearningProject

Environment creation

This creates the virtual enviroment within a directory env.

python -m venv env

Then you can activate the environment.

source env/bin/activate

Alternatively, in case of fish shell.

source env/bin/activate.fish

Install dependencies

Automatically install all required packages.

pip install -r requirements.txt

Download the dataset

There are two ways for importing the dataset.

A. Construct the dataset from scratch (not deterministic in IID)

Create dataset directory

cd data/femnist

Run the following commands to generate the datasets (and download the necessary files if missing)

To generate the non-iid distribution

./preprocess_parquet.sh -s niid --sf 1.0 -k 0 -t sample

To generate the iid distribution

./preprocess_parquet.sh -s iid --sf 1.0 -k 0 -t sample

Go back to parent directory of the project

cd -

B. Use the constructed dataset (deterministic for shared usage)

Download zipped file.

wget --load-cookies /tmp/cookies.txt "https://docs.google.com/uc?export=download&confirm=$(wget --quiet --save-cookies /tmp/cookies.txt --keep-session-cookies --no-check-certificate 'https://docs.google.com/uc?export=download&id=1AXJ5uuswGkv9dzVGMAJGRbGViWZFgImE' -O- | sed -rn 's/.*confirm=([0-9A-Za-z_]+).*/\1\n/p')&id=1AXJ5uuswGkv9dzVGMAJGRbGViWZFgImE" -O compressed.zip && rm -rf /tmp/cookies.txt

Unzip file and remove zipped dataset.

unzip compressed.zip -d data/femnist/ 
rm -rf compressed.zip

Access to Weights & Biases

Execute the following command to login with API key b578cc4325e4b0652255efe8f2878be1d5fad2f2.

wandb login b578cc4325e4b0652255efe8f2878be1d5fad2f2

This will provide automatic access and log to the Weights & Biases platform for model and experiments tracking.

Run the application

Run the application and test the model.

python main.py --niid --seed 0 --dataset femnist --model cnn \
    --num_rounds 1000 \
    --num_epochs 1 \
    --clients_per_round 10 \
    --print_train_interval 1000 --print_test_interval 1000 \
    --eval_interval 10 --test_interval 10

By tweaking the parameters, we can have multiple experiments and outcomes.