tracerec

Python package for the creation of recommendation systems based on user traces.

Inslation

With pip:

pip install tracerec

With Poetry:

poetry add tracerec

Development

For contributing to the development of this package, you can clone the repository and install the dependencies using Poetry.

# Clone the repository
git clone https://github.com/almtav08/tracerec.git
cd tracerec

# Install dev dependencies
poetry install

Usage

import torch
from tracerec.algorithms.knowledge_based.transe import TransE
from tracerec.data.triples.triples_manager import TriplesManager
from tracerec.samplers.path_based_sampler import PathBasedNegativeSampler


# Create a sample triples manager with some triples
triples = [
    (1, 0, 2),
    (2, 0, 3),
    (3, 0, 4)
]
triples_manager = TriplesManager(triples)

train_x, train_y, test_x, test_y = triples_manager.split(train_ratio=0.8, relation_ratio=True, random_state=42, device='cpu')

# Negative sampling
all_triples = triples_manager.get_triples()
all_entities = triples_manager.get_entities()
entity_paths = triples_manager.get_entity_paths()

sampler = PathBasedNegativeSampler(all_triples, all_entities, corruption_ration=0.5, device='cpu', entity_paths=entity_paths, min_distance=1.0)
train_x_neg = sampler.sample(train_x, num_samples=1, random_state=42)

# Create and compile the TransE model
transe = TransE(num_entities=4, num_relations=1, embedding_dim=10, device='cpu', norm=1)
transe.compile(optimizer=torch.optim.Adam, criterion=torch.nn.MarginRankingLoss(margin=1.0))

# Fit the model
transe.fit(train_x, train_x_neg, train_y, num_epochs=1, batch_size=1, lr=0.001, verbose=True)

License

MIT