Quickstart
flwr new @hackathon/berlin25-eurosatReadme
EuroSAT — Flower / PyTorch App (Track 2)
This app demonstrates federated learning on the EuroSAT remote-sensing dataset using Flower and PyTorch.
It’s designed to help participants explore and understand how the dataset and federated setup work.
🚀 Quickstart
1) Install
First, install dependencies in editable mode so that changes you make to the app code are immediately reflected in your environment.
pip install -e .2) Run (Simulation Engine)
You can run the app using Flower’s Simulation Runtime, which lets you simulate multiple clients locally.
This is the easiest way to experiment and debug your app before scaling it to real devices.
Run with default configuration:
flwr run .Tip: Your pyproject.toml file can define more than just dependencies — it can also include hyperparameters (like lr or num-server-rounds) and control which Flower Runtime is used.
By default, this app uses the Simulation Runtime, but you can switch to the Deployment Runtime when needed.
Learn more in the TOML configuration guide.
🌸 Explore Flower Datasets
The code below shows how to load, transform, and inspect the EuroSAT dataset using flwr_datasets.
Each client receives a separate partition of the dataset which is to simulate data distributed across multiple devices.
from flwr_datasets import FederatedDataset
from torchvision import transforms
# Define a simple transform for a single image
img_transform = transforms.Compose([
transforms.Resize((64, 64)),
transforms.ToTensor(),
])
# Create a federated dataset with 10 IID partitions
fds = FederatedDataset(dataset="tanganke/eurosat", partitioners={"train": 10})
# Load one partition for local exploration
partition = fds.load_partition(0, split="train")
# HF Datasets' with_transform gets a *batch* dict: each value is a list
def hf_transform(batch):
# Apply the torchvision transform to each image in the batch
batch["image"] = [img_transform(img) for img in batch["image"]]
return batch
partition = partition.with_transform(hf_transform)
# Inspect a few samples
for i in range(3):
sample = partition[i]
print(sample["image"].shape, sample["label"])💡 What this does:
- Loads EuroSAT (RGB version) from Hugging Face using Flower’s dataset abstraction.
- Partitions it into 10 federated clients (simulated).
- Applies a transformation (resize + tensor) to images.
- Prints out a few samples so you can confirm everything works before training.
📊 Visualize Label Distributions
Understanding how data is distributed across clients is key to federated learning — it helps explain differences in model performance and convergence.
In the following, we can observe how to load the dataset, apply a transformation to each image, and then visualize how the class labels are distributed across all client partitions. It also allows plotting the label distribution for a single client to inspect the local data balance.
from flwr_datasets import FederatedDataset
from flwr_datasets.visualization import plot_label_distributions
import matplotlib.pyplot as plt
fds = FederatedDataset(
dataset="tanganke/eurosat",
partitioners={"train": 10},
)
# Get the partitioner used for "train"
partitioner = fds.partitioners["train"]
# Plot label distribution across *all* 10 partitions
fig, ax, df = plot_label_distributions(
partitioner=partitioner,
label_name="label",
plot_type="bar",
size_unit="percent",
partition_id_axis="x",
legend=True,
verbose_labels=False,
title="EuroSAT Label Distribution per Client (train)",
)
plt.show()
print(df.head())💡 What this does:
- Uses Flower’s built-in plot_label_distributions utility to visualize the dataset split across clients.
- Quickly shows whether partitions are IID or skewed.
- Returns a pandas DataFrame (df) so you can further analyze counts or percentages per class.
Run with the Deployment Engine
Follow this how-to guide to run the same app in this example but with Flower's Deployment Engine. After that, you might be interested in setting up secure TLS-enabled communications and SuperNode authentication in your federation.
You can run Flower on Docker too! Check out the Flower with Docker documentation.