Cornac is a comparative framework for multimodal recommender systems. It focuses on making it convenient to work with models leveraging auxiliary data (e.g., item descriptive text and image, social network, etc). Cornac enables fast experiments and straightforward implementations of new models. It is highly compatible with existing machine learning libraries (e.g., TensorFlow, PyTorch).
Cornac is one of the frameworks recommended by ACM RecSys 2023 for the evaluation and reproducibility of recommendation algorithms.
Website | Documentation | Tutorials | Examples | Models | Datasets | Paper | Preferred.AI
Currently, we are supporting Python 3. There are several ways to install Cornac:
From PyPI (recommended):
pip3 install cornac
From Anaconda:
conda install cornac -c conda-forge
From the GitHub source (for latest updates):
pip3 install Cython numpy scipy
pip3 install git+https://github.com/PreferredAI/cornac.git
Note:
Additional dependencies required by models are listed here.
Some algorithm implementations use OpenMP to support multi-threading. For Mac OS users, in order to run those algorithms efficiently, you might need to install gcc from Homebrew to have an OpenMP compiler:
brew install gcc | brew link gcc
Flow of an Experiment in Cornac
import cornac
from cornac.eval_methods import RatioSplit
from cornac.models import MF, PMF, BPR
from cornac.metrics import MAE, RMSE, Precision, Recall, NDCG, AUC, MAP
# load the built-in MovieLens 100K and split the data based on ratio
ml_100k = cornac.datasets.movielens.load_feedback()
rs = RatioSplit(data=ml_100k, test_size=0.2, rating_threshold=4.0, seed=123)
# initialize models, here we are comparing: Biased MF, PMF, and BPR
mf = MF(k=10, max_iter=25, learning_rate=0.01, lambda_reg=0.02, use_bias=True, seed=123)
pmf = PMF(k=10, max_iter=100, learning_rate=0.001, lambda_reg=0.001, seed=123)
bpr = BPR(k=10, max_iter=200, learning_rate=0.001, lambda_reg=0.01, seed=123)
models = [mf, pmf, bpr]
# define metrics to evaluate the models
metrics = [MAE(), RMSE(), Precision(k=10), Recall(k=10), NDCG(k=10), AUC(), MAP()]
# put it together in an experiment, voilà!
cornac.Experiment(eval_method=rs, models=models, metrics=metrics, user_based=True).run()
Output:
| MAE | RMSE | AUC | MAP | NDCG@10 | Precision@10 | Recall@10 | Train (s) | Test (s) | |
|---|---|---|---|---|---|---|---|---|---|
| MF | 0.7430 | 0.8998 | 0.7445 | 0.0548 | 0.0761 | 0.0675 | 0.0463 | 0.13 | 1.57 |
| PMF | 0.7534 | 0.9138 | 0.7744 | 0.0671 | 0.0969 | 0.0813 | 0.0639 | 2.18 | 1.64 |
| BPR | N/A | N/A | 0.8695 | 0.1042 | 0.1500 | 0.1110 | 0.1195 | 3.74 | 1.49 |
Here, we provide a simple way to serve a Cornac model by launching a standalone web service with Flask. It is very handy for testing or creating a demo application. First, we install the dependency:
$ pip3 install Flask
Supposed that we want to serve the trained BPR model from previous example, we need to save it:
bpr.save("save_dir", save_trainset=True)
After that, the model can be deployed easily by running Cornac serving app as follows:
$ FLASK_APP='cornac.serving.app' \
MODEL_PATH='save_dir/BPR' \
MODEL_CLASS='cornac.models.BPR' \
flask run --host localhost --port 8080
# Running on http://localhost:8080
Here we go, our model service is now ready. Let's get top-5 item recommendations for the user "63":
$ curl -X GET "http://localhost:8080/recommend?uid=63&k=5&remove_seen=false"
# Response: {"recommendations": ["50", "181", "100", "258", "286"], "query": {"uid": "63", "k": 5, "remove_seen": false}}
If we want to remove seen items during training, we need to provide TRAIN_SET which has been saved with the model earlier, when starting the serving app. We can also leverage WSGI server for model deployment in production. Please refer to this guide for more details.
One important aspect of deploying recommender model is efficient retrieval via Approximate Nearest Neighbor (ANN) search in vector space. Cornac integrates several vector similarity search frameworks for the ease of deployment. This example demonstrates how ANN search will work seamlessly with any recommender models supporting it (e.g., matrix factorization).
| Supported Framework | Cornac Wrapper | Example |
|---|---|---|
| spotify/annoy | AnnoyANN | quick-start, deep-dive |
| meta/faiss | FaissANN | quick-start, deep-dive |
| nmslib/hnswlib | HNSWLibANN | quick-start, hnsw-lib, deep-dive |
| google/scann | ScaNNANN | quick-start, deep-dive |
The table below lists the recommendation models/algorithms featured in Cornac. Examples are provided as quick-start showcasing an easy to run script, or as deep-dive explaining the math and intuition behind each model. Why don't you join us to lengthen the list?
This project welcomes contributions and suggestions. Before contributing, please see our contribution guidelines.
If you use Cornac in a scientific publication, we would appreciate citations to the following papers:
@article{salah2020cornac,
title={Cornac: A Comparative Framework for Multimodal Recommender Systems},
author={Salah, Aghiles and Truong, Quoc-Tuan and Lauw, Hady W},
journal={Journal of Machine Learning Research},
volume={21},
number={95},
pages={1--5},
year={2020}
}
@article{truong2021exploring,
title={Exploring Cross-Modality Utilization in Recommender Systems},
author={Truong, Quoc-Tuan and Salah, Aghiles and Tran, Thanh-Binh and Guo, Jingyao and Lauw, Hady W},
journal={IEEE Internet Computing},
year={2021},
publisher={IEEE}
}
@inproceedings{truong2021multi,
title={Multi-modal recommender systems: Hands-on exploration},
author={Truong, Quoc-Tuan and Salah, Aghiles and Lauw, Hady},
booktitle={Fifteenth ACM Conference on Recommender Systems},
pages={834--837},
year={2021}
}
FAQs
A Comparative Framework for Multimodal Recommender Systems
We found that cornac demonstrated a healthy version release cadence and project activity because the last version was released less than a year ago. It has 1 open source maintainer collaborating on the project.
Did you know?
Socket for GitHub automatically highlights issues in each pull request and monitors the health of all your open source dependencies. Discover the contents of your packages and block harmful activity before you install or update your dependencies.
Security News
Elastic’s return to open source with the AGPL license has been met with skepticism, as many developers see it as a strategic move rather than a genuine effort to restore user trust and freedoms.
Security News
A new "revival hijack" supply chain attack targets deleted Python packages, with an estimated 22K packages at risk. Socket can detect and block hijacked packages that have added malicious code.
Product
We're introducing a new Analytics feature in the Socket dashboard so you can view changes in your organization's and repositories' alerts over time.