Quick Start with PLANETALIGN

This guide provides a minimal working example for running a built-in network alignment (NA) method on a built-in dataset using PLANETALIGN.

Installation 

Install from source by downloading from the anonymous repository and building locally:

cd PlanetAlign-E9BA
pip install -e .

Basic Usage 

Here is a minimal example of aligning two social networks (Douban dataset) using the FINAL algorithm:

import PlanetAlign

# Download and load the Douban dataset
dataset = PlanetAlign.datasets.Douban(
    root='data/',
    download=True,
    train_ratio=0.2,
    seed=42
)

# Initialize the FINAL alignment model
model = PlanetAlign.algorithms.FINAL(
    alpha=0.9,  # hyperparameter specific to FINAL
).to('cuda')  # or 'cpu'

# Initialize logger
logger = PlanetAlign.logger.TrainLogger(
    log_dir='logs/',
    log_name='final_douban',
    save=True
)

# Train the model
model.train(
    dataset=dataset,
    gid1=0,          # index of the first graph
    gid2=1,          # index of the second graph
    use_attr=True,      # use attributes if available
    logger=logger,
    total_epochs=50
)

# Evaluate the model
result = model.test(
    dataset=dataset,
    gids=[0, 1],
    metrics=['Hits@1', 'Hits@10', 'MRR']
)

print(result)

Visualizing Training Metrics 

After training, you can visualize metrics like training loss or memory usage:

logger.plot_curve(metric='Hits@1', save_path='plots/hits1.png')

Next Steps 

Explore other datasets: FoursquareTwitter, PhoneEmail, ACM_DBLP, etc.
Try other algorithms: JOENA, PARROT, NeXtAlign, etc.
Define your own dataset or model by inheriting from Dataset or BaseModel.