mlvern

mlvern is a Python library for structuring machine learning workflows with consistent dataset handling, experiment tracking, and model management.

It provides a lightweight framework to organize ML projects by separating data processing, experimentation, and evaluation into reproducible units.

https://ml-vern.readthedocs.io/en/latest/

• Dataset registration with fingerprint-based identification
• Metadata tracking for datasets and experiments
• Structured experiment execution workflow
• Model artifact storage and retrieval
• Evaluation tracking and comparison across runs
• Simple prediction interface for trained models
• Utilities for dataset inspection and validation

mlvern is built around the following principles:

• Reproducibility: identical inputs produce identical tracked outputs
• Traceability: datasets, experiments, and models are versioned and linked
• Simplicity: minimal API surface with explicit behavior
• Separation of concerns: data, training, and evaluation are decoupled
• Lightweight structure: avoids unnecessary abstraction layers

pip install mlvern

from mlvern import Forge

forge = Forge("your_project", "your_base_dir")
forge.init()
dataset_fp, _ = forge.register_dataset(df, "target")
run_id, metrics = forge.run(model, X_train, y_train, X_val, y_val, config, dataset_fp)

from mlvern import ModelComparator
ModelComparator(forge).compare_models([run_id])

Python 3.8+, NumPy, Pandas