vmware-samples / efficient-multiclass-classification

Duet is a scikit-learn classifier for resource-efficient multiclass classification that incorporates the advantages of bagging and boosting decision-tree-based ensemble methods (DTEMs) by using two classifiers instead of a monolithic one. A simple bagging model is trained using the entire training dataset and is responsible for capturing the easier concepts. Then, a boosting model is trained using only a fraction of the dataset representing the concepts the bagging model finds hard.

Duet scikit classifier (v1.0)

Overview

Duet is a decision tree ensemble method based multiclass classification framework that offers a more efficient resource usage while preserving and even improving the classification accuracy in comparison to standard monolithic models.

Duet is based on a small bagging ensemble model and a booting model.
The current implementation of Duet is based on Random Forest and XGBoost.

Documentation

More details about the Duet can be found in the following paper:
"Efficient Multiclass Classification with Duet" [EuroMLSys '22]
https://dl.acm.org/doi/abs/10.1145/3517207.3526970
https://euromlsys.eu/pdf/euromlsys22-final4.pdf

Files:

duet_classifier.py

Duet scikit classifier

classification_example.py

Basic classification example by Duet

grid_search_example.py

Basic grid search example with Duet

Prerequisities:

numpy
pandas
skleran
xgboost

or alternatively, run:
$ pip3 install -r requirements.txt

Contributing

The efficient-multiclass-classification project team welcomes contributions from the community. Before you start working with efficient-multiclass-classification, please read our Developer Certificate of Origin. All contributions to this repository must be signed as described on that page. Your signature certifies that you wrote the patch or have the right to pass it on as an open-source patch. For more detailed information, refer to CONTRIBUTING.md.

License

BSD-3 License

Contact us

For more information, support and advanced examples contact:
Yaniv Ben-Itzhak, [email protected]
Shay Vargaftik, [email protected]