Jarvis is a toolbox built on top of TensorFlow2.0 that allows developers and researchers to easily build neural networks in TensorFlow, particularly CTR models for large-scale advertising and recommendation scenarios. It provides the implementation of Meitu's FLEN model.
Note that Jarvis is still actively under development, so feedback and contributions are welcome. Feel free to submit your contributions as a pull request.
- Scalability: fast training on large-scale networks with tens of millions of sparse features
- Extensible: easily register new models and criteria.
- Supported tasks:
- CTR prediction
- Multi-task learning (coming)
- online learning (todo)
Requirements and Installation
Please see environment.yml for more details
You can use
python scripts/flen.py to run FLEN model on Avazu dataset.
Download the tfrecord format dataset from here.
Alternatively, You can use
python tools/dataset/avazu.py to prepare Avazu dataset yourself.
Implement Your Own Model
If you have a well-perform algorithm and are willing to implement it in our toolkit to help more people, you can create a pull request, detailed information can be found here.