Example: PyTorch - From Centralized To Federated

This tutorial will show you how to use Flower to build a federated version of an existing machine learning workload. We are using PyTorch to train a Convolutional Neural Network on the CIFAR-10 dataset. First, we introduce this machine learning task with a centralized training approach based on the Deep Learning with PyTorch tutorial. Then, we build upon the centralized training code to run the training in a federated fashion.

Centralized Training

We begin with a brief description of the centralized CNN training code. If you want a more in-depth explanation of what’s going on then have a look at the official PyTorch tutorial.

Let’s create a new file called cifar.py with all the components required for a traditional (centralized) training on CIFAR-10. First, all required packages (such as torch and torchvision) need to be imported. You can see that we do not import any package for federated learning. You can keep all these imports as they are even when we add the federated learning components at a later point.

from typing import Tuple, Dict

import torch
import torch.nn as nn
import torch.nn.functional as F
import torchvision
import torchvision.transforms as transforms
from torch import Tensor
from torchvision.datasets import CIFAR10

As already mentioned we will use the CIFAR-10 dataset for this machine learning workload. The model architecture (a very simple Convolutional Neural Network) is defined in class Net().

class Net(nn.Module):

    def __init__(self) -> None:
        super(Net, self).__init__()
        self.conv1 = nn.Conv2d(3, 6, 5)
        self.pool = nn.MaxPool2d(2, 2)
        self.conv2 = nn.Conv2d(6, 16, 5)
        self.fc1 = nn.Linear(16 * 5 * 5, 120)
        self.fc2 = nn.Linear(120, 84)
        self.fc3 = nn.Linear(84, 10)

    def forward(self, x: Tensor) -> Tensor:
        x = self.pool(F.relu(self.conv1(x)))
        x = self.pool(F.relu(self.conv2(x)))
        x = x.view(-1, 16 * 5 * 5)
        x = F.relu(self.fc1(x))
        x = F.relu(self.fc2(x))
        x = self.fc3(x)
        return x

The load_data() function loads the CIFAR-10 training and test sets. The transform normalized the data after loading.

DATA_ROOT = "~/data/cifar-10"


def load_data() -> (
    Tuple[torch.utils.data.DataLoader, torch.utils.data.DataLoader, Dict]
):
    """Load CIFAR-10 (training and test set)."""
    transform = transforms.Compose(
        [transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))]
    )
    trainset = CIFAR10(DATA_ROOT, train=True, download=True, transform=transform)
    trainloader = torch.utils.data.DataLoader(trainset, batch_size=32, shuffle=True)
    testset = CIFAR10(DATA_ROOT, train=False, download=True, transform=transform)
    testloader = torch.utils.data.DataLoader(testset, batch_size=32, shuffle=False)
    num_examples = {"trainset": len(trainset), "testset": len(testset)}
    return trainloader, testloader, num_examples

We now need to define the training (function train()) which loops over the training set, measures the loss, backpropagates it, and then takes one optimizer step for each batch of training examples.

The evaluation of the model is defined in the function test(). The function loops over all test samples and measures the loss of the model based on the test dataset.

def train(
    net: Net,
    trainloader: torch.utils.data.DataLoader,
    epochs: int,
    device: torch.device,
) -> None:
    """Train the network."""
    # Define loss and optimizer
    criterion = nn.CrossEntropyLoss()
    optimizer = torch.optim.SGD(net.parameters(), lr=0.001, momentum=0.9)

    print(f"Training {epochs} epoch(s) w/ {len(trainloader)} batches each")

    # Train the network
    for epoch in range(epochs):  # loop over the dataset multiple times
        running_loss = 0.0
        for i, data in enumerate(trainloader, 0):
            images, labels = data[0].to(device), data[1].to(device)

            # zero the parameter gradients
            optimizer.zero_grad()

            # forward + backward + optimize
            outputs = net(images)
            loss = criterion(outputs, labels)
            loss.backward()
            optimizer.step()

            # print statistics
            running_loss += loss.item()
            if i % 100 == 99:  # print every 100 mini-batches
                print("[%d, %5d] loss: %.3f" % (epoch + 1, i + 1, running_loss / 2000))
                running_loss = 0.0


def test(
    net: Net,
    testloader: torch.utils.data.DataLoader,
    device: torch.device,
) -> Tuple[float, float]:
    """Validate the network on the entire test set."""
    criterion = nn.CrossEntropyLoss()
    correct = 0
    total = 0
    loss = 0.0
    with torch.no_grad():
        for data in testloader:
            images, labels = data[0].to(device), data[1].to(device)
            outputs = net(images)
            loss += criterion(outputs, labels).item()
            _, predicted = torch.max(outputs.data, 1)
            total += labels.size(0)
            correct += (predicted == labels).sum().item()
    accuracy = correct / total
    return loss, accuracy

Having defined the data loading, model architecture, training, and evaluation we can put everything together and train our CNN on CIFAR-10.

def main():
    DEVICE = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
    print("Centralized PyTorch training")
    print("Load data")
    trainloader, testloader, _ = load_data()
    print("Start training")
    net = Net().to(DEVICE)
    train(net=net, trainloader=trainloader, epochs=2, device=DEVICE)
    print("Evaluate model")
    loss, accuracy = test(net=net, testloader=testloader, device=DEVICE)
    print("Loss: ", loss)
    print("Accuracy: ", accuracy)


if __name__ == "__main__":
    main()

You can now run your machine learning workload:

python3 cifar.py

So far, this should all look fairly familiar if you’ve used PyTorch before. Let’s take the next step and use what we’ve built to create a simple federated learning system consisting of one server and two clients.

Federated Training

The simple machine learning project discussed in the previous section trains the model on a single dataset (CIFAR-10), we call this centralized learning. This concept of centralized learning, as shown in the previous section, is probably known to most of you, and many of you have used it previously. Normally, if you’d want to run machine learning workloads in a federated fashion, then you’d have to change most of your code and set everything up from scratch. This can be a considerable effort.

However, with Flower you can evolve your pre-existing code into a federated learning setup without the need for a major rewrite.

The concept is easy to understand. We have to start a server and then use the code in cifar.py for the clients that are connected to the server. The server sends model parameters to the clients. The clients run the training and update the parameters. The updated parameters are sent back to the server which averages all received parameter updates. This describes one round of the federated learning process and we repeat this for multiple rounds.

Our example consists of one server and two clients. Let’s set up server.py first. The server needs to import the Flower package flwr. Next, we use the start_server function to start a server and tell it to perform three rounds of federated learning.

import flwr as fl

if __name__ == "__main__":
    fl.server.start_server(
        server_address="0.0.0.0:8080", config=fl.server.ServerConfig(num_rounds=3)
    )

We can already start the server:

python3 server.py

Finally, we will define our client logic in client.py and build upon the previously defined centralized training in cifar.py. Our client needs to import flwr, but also torch to update the parameters on our PyTorch model:

from collections import OrderedDict
from typing import Dict, List, Tuple

import numpy as np
import torch

import cifar
import flwr as fl

DEVICE: str = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

Implementing a Flower client basically means implementing a subclass of either flwr.client.Client or flwr.client.NumPyClient. Our implementation will be based on flwr.client.NumPyClient and we’ll call it CifarClient. NumPyClient is slightly easier to implement than Client if you use a framework with good NumPy interoperability (like PyTorch or TensorFlow/Keras) because it avoids some of the boilerplate that would otherwise be necessary. CifarClient needs to implement four methods, two methods for getting/setting model parameters, one method for training the model, and one method for testing the model:

1. set_parameters

   • set the model parameters on the local model that are received from the server

   • loop over the list of model parameters received as NumPy ndarray’s (think list of neural network layers)

2. get_parameters

   • get the model parameters and return them as a list of NumPy ndarray’s (which is what flwr.client.NumPyClient expects)

3. fit

   • update the parameters of the local model with the parameters received from the server

   • train the model on the local training set

   • get the updated local model weights and return them to the server

4. evaluate

   • update the parameters of the local model with the parameters received from the server

   • evaluate the updated model on the local test set

   • return the local loss and accuracy to the server

The two NumPyClient methods fit and evaluate make use of the functions train() and test() previously defined in cifar.py. So what we really do here is we tell Flower through our NumPyClient subclass which of our already defined functions to call for training and evaluation. We included type annotations to give you a better understanding of the data types that get passed around.

class CifarClient(fl.client.NumPyClient):
    """Flower client implementing CIFAR-10 image classification using
    PyTorch."""

    def __init__(
        self,
        model: cifar.Net,
        trainloader: torch.utils.data.DataLoader,
        testloader: torch.utils.data.DataLoader,
        num_examples: Dict,
    ) -> None:
        self.model = model
        self.trainloader = trainloader
        self.testloader = testloader
        self.num_examples = num_examples

    def get_parameters(self, config) -> List[np.ndarray]:
        # Return model parameters as a list of NumPy ndarrays
        return [val.cpu().numpy() for _, val in self.model.state_dict().items()]

    def set_parameters(self, parameters: List[np.ndarray]) -> None:
        # Set model parameters from a list of NumPy ndarrays
        params_dict = zip(self.model.state_dict().keys(), parameters)
        state_dict = OrderedDict({k: torch.tensor(v) for k, v in params_dict})
        self.model.load_state_dict(state_dict, strict=True)

    def fit(
        self, parameters: List[np.ndarray], config: Dict[str, str]
    ) -> Tuple[List[np.ndarray], int, Dict]:
        # Set model parameters, train model, return updated model parameters
        self.set_parameters(parameters)
        cifar.train(self.model, self.trainloader, epochs=1, device=DEVICE)
        return self.get_parameters(config={}), self.num_examples["trainset"], {}

    def evaluate(
        self, parameters: List[np.ndarray], config: Dict[str, str]
    ) -> Tuple[float, int, Dict]:
        # Set model parameters, evaluate model on local test dataset, return result
        self.set_parameters(parameters)
        loss, accuracy = cifar.test(self.model, self.testloader, device=DEVICE)
        return float(loss), self.num_examples["testset"], {"accuracy": float(accuracy)}

All that’s left to do it to define a function that loads both model and data, creates a CifarClient, and starts this client. You load your data and model by using cifar.py. Start CifarClient with the function fl.client.start_client() by pointing it at the same IP address we used in server.py:

def main() -> None:
    """Load data, start CifarClient."""

    # Load model and data
    model = cifar.Net()
    model.to(DEVICE)
    trainloader, testloader, num_examples = cifar.load_data()

    # Start client
    client = CifarClient(model, trainloader, testloader, num_examples)
    fl.client.start_client(server_address="0.0.0.0:8080", client.to_client())


if __name__ == "__main__":
    main()

And that’s it. You can now open two additional terminal windows and run

python3 client.py

in each window (make sure that the server is running before you do so) and see your (previously centralized) PyTorch project run federated learning across two clients. Congratulations!

Next Steps

The full source code for this example: PyTorch: From Centralized To Federated (Code). Our example is, of course, somewhat over-simplified because both clients load the exact same dataset, which isn’t realistic. You’re now prepared to explore this topic further. How about using different subsets of CIFAR-10 on each client? How about adding more clients?