Scorta

Scorta is a versatile library designed to facilitate the reuse of machine learning pipelines across multiple projects, including but not limited to 2-stage recommender systems. It offers a streamlined and straightforward interface that can be easily adapted for various machine learning competitions.

Features

• Reusability: Crafted with the aim to be reused in multiple ML pipelines.
• Compatibility: Suitable for use in 2-stage recommender systems and more.
• Examples Included: The samples directory contains real-world code examples from ML competitions.
• Simplicity and Flexibility: Maintains simplicity while being adaptable for use in different competition contexts.

Installation

You can install Scorta directly from the repository using pip:

pip3 install git+https://github.com/zerebom/scorta.git

Directory Structure

Scorta comes with a well-organized directory structure:

scorta/
├── eda/
├── feature/
├── model/
├── recsys/
├── testing/
└── utils/
samples/
├── 001_binary_classification.ipynb
├── atma-16/
└── atma15/

Within the samples directory, you can find notebooks (*.ipynb) demonstrating how scorta was utilized in ML competitions, providing a practical reference for your projects.

Usage Example

Here's a quick example of how to extend the Candidate class from the scorta.recsys module for a competition:

from scorta.recsys.candidate_generate import Candidate

class Atma16Candidate(Candidate):
    # initialization and method definitions
    ...

class LastSeenedCandidate(Atma16Candidate):
    # A specific strategy for scoring within the same session
    def generate(self) -> pl.DataFrame:
        # Logic to generate candidate DataFrame
        ...

With Scorta, you can define your own candidates and strategies to fit the needs of your recommender system, ensuring high reusability and easy experimentation.

Contributing

Contributions are what make the open-source community such an amazing place to learn, inspire, and create. Any contributions you make are greatly appreciated.

Contact

If you have any questions or suggestions, please reach out to us at the repository on GitHub.