Skip to main content

Data Versioning

Updated May 13, 2023 ·

Overview​

Tracking data and model versions is essential in machine learning to ensure reproducibility and traceability.

  • Easy tracking and rollback of changes to data and models.
  • Differentiates large updates from small tweaks.
  • Helps identify which data was used for experiments.
  • Ensures that model changes are tracked alongside data.

Types of Versioning​

Versioning is categorized into two types:

  • Major Versioning:

    • Represents significant changes
    • Example: Adding new features or methods

  • Minor Versioning

    • Indicates smaller updates
    • Bug fixes or slight adjustments.

Versioning Training Data​

We can version data by labeling it with unique identifiers. For example:

  • Version 1.0: Initial dataset.
  • Version 1.1: Added data transformations (e.g., scaling).
  • Version 1.2: Added feature selection (e.g., Chi-squared).
  • Version 2.0: Major update with new data.

Feature Store​

A feature store manages the versions of features across experiments, ensures consistent data usage in future tests, and avoids duplication.

For more information, please see Feature Store.

Versioning ML Models​

Model versioning allows tracking of model changes over time. For example:

  • Model version 1.0: Initial model trained with data version 1.0.
  • Model version 1.1: Fine-tuned model with data version 1.1.
  • Model version 2.0: New model (e.g., XGBoost) trained with data version 1.2.

MLflow for Model Versioning​

Here’s how you can log a model version with MLflow:

import mlflow

# Set version to 1.0
mlflow.set_tag("model_version", "1.0")

# Log model
with mlflow.start_run():
    mlflow.log_model(model, "model")

This logs the model version, and you can track changes across experiments.

Model Stores​

A model store is like a feature store but for models. It ensures models can be reused and reproduced across experiments.

  • Enable version control and rollback
  • Works well in tandem with feature stores