Skip to main content

Introduction

What is the TANGO Library?​

Formal definitions​

TANGO AI Methods: novel algorithms and methods developed within WP2 and WP3

TANGO AI Methods implementation: software implementation of the algorithms and methods devised in WP2 and WP3:

  • some of it may be released as open source on a public repository (github or the alike);
  • some of it may become integrated within the TANGO library.

TANGO Library: The TANGO Library is the project’s open-source, curated repository of methods, algorithms and utilities that enable TANGO models to be hosted in the TANGO Platform. Implemented in Python and released under the Apache v2 license, public and fully documented, it is designed to:

  • provide shared interfaces, classes and utilities designed for reusability and smooth integration within the TANGO Ecosystem;
  • host methods (from WP2/3 contributors) intended for direct usage within TANGO’s workflows;
  • facilitate End-to-End Application Building: the core library acts as the “contract” for implementing TANGO’s.

Components​

Core: the foundational building blocks of the TANGO Ecosystem: TANGO interfaces, enumerations, data-model, connectors implementations (as of today, mlflow & databricks)

Utils: reusable functions or small libraries intended to simplify development (e.g. generate explanations for model outputs or handle common tasks like data preprocessing or validation); some of them may come from WP2/WP3

Model Code: the machine learning codebase that implements one or more of the TANGO Core interfaces (e.g., for data preprocessing, model training, inference, or explanation). This code is responsible for:

  • defining how data is prepared (pre-processing) and how model outputs are interpreted (post-processing);
  • training new model instances, including handling hyperparameters, checkpoints and local storage of relevant artifacts;
  • providing model outcome to invocations: once deployed on the target MLOps, the trained artifact can then be registered, deployed and served via the TANGO platform.