SnapperML

SnapperML is a comprehensive framework for experiment tracking and machine learning operationalization (MLOps), built using well-supported technologies like Mlflow, Ray, Docker, and more. It provides an opinionated workflow designed to facilitate both local and cloud-based experimentation.

Key Features

• Automatic Tracking: Seamless integration with MLflow for parameter and metric tracking.
• Distributed Training: First-class support for distributed training and hyperparameter optimization using Optuna and Ray.
• CLI-Based Execution: Easily package and execute projects within containers using our intuitive Command Line Interface (CLI).
• Web Interface: A modern web interface developed with Vite, React, TypeScript, and Bootstrap for managing experiment configurations.

Project Goals

SnapperML aims to:

1. Enhance Maintainability: By addressing technical debt and improving the codebase, making it cleaner and more efficient.
2. Improve Scalability: Ensure the system can handle large-scale experiments and concurrent requests smoothly.
3. Provide a Robust Web UI: A user-friendly interface that simplifies the setup and execution of ML experiments.
4. Ensure Reproducibility: Leverage MLOps principles to ensure experiments can be replicated easily.

Architecture

Overview

SnapperML integrates several components to streamline machine learning workflows:

• CLI Framework: Facilitates command-based interactions and logging for experiment execution.
• Flask API: Manages requests from the frontend and interfaces with backend processes.
• Vite-Powered Web UI: An accessible and intuitive web application that handles experiment configurations and tracks real-time logs.
• Containerized Databases: Securely stores experiment results using containerized MLflow and Optuna databases.

Installation

Prerequisites

• docker
• python 3.12+
• node.js (for UI development)

Install

The python package can be install using pip:

pip install snapper-ml

Or from this repo:

pip install .

Deploy

To run SnapperML, you first need to deploy MLflow and Optuna databases. Execute:

snapper-ml make docker

Once the deploy finished you can execute snapper-ml in the CLI. For an ilustrative example, check the example section.

To use snapperML web interface you need to deploy it too.

snapper-ml make UI

Open localhost:4000 and upload your firsts experiments!

To stop snapper UI just execute:

make stop_UI

And to stop mlflow and optuna databases execute:

make stop_docker

Documentation

The documentation is available here

Example

# train_svm.py

from snapper_ml import job

@job
def main(C, kernel, gamma='scale'):
    np.random.seed(1234)
    X_train, X_val, y_train, y_val = load_data()
    model = SVC(C=C, gamma=gamma, kernel=kernel)
    model.fit(X_train, y_train)
    accuracy = model.score(X_val, y_val)
    return {'val_accuracy': accuracy}


if __name__ == '__main__':
    main()

# train_svm.yaml

name: "SVM"
kind: "group"
num_trials: 12
sampler: TPE

param_space:
  C: loguniform(0.01, 1000)
  gamma: choice(['scale', 'auto'])

metric:
  name: val_accuracy
  direction: maximize

ray_config:
  num_cpus: 4

data:
  folder: data/
  files: ["*QGSJet.txt"]

run:
  - train_svm.py

snapper-ml run --config_file=train_svm.yaml

There are more examples in the examples folder.