Getting Started

Minerva is a framework that provides tools and libraries for developing machine learning models, particularly in seismic and time-series processing, as well as self-supervised learning tasks. It also includes utilities for conducting reproducible experiments.

Built on top of PyTorch and PyTorch Lightning, Minerva is designed to be modular and extensible, allowing users to integrate and extend its functionalities with ease.

Once installed, Minerva functions like any other Python package. You can import its modules and classes as follows:

import minerva

Core Modules

Minerva consists of several specialized modules, each supporting different aspects of model development:

Module	Description
minerva.analysis	Tools for analyzing and visualizing model performance.
minerva.callback	Callbacks for monitoring and logging training progress (built on PyTorch Lightning).
minerva.data	Readers, datasets, and data modules.
minerva.losses	Collection of loss functions.
minerva.models	Predefined models and architectures for supervised and self-supervised learning. For more details, check the minerva.models README.md.
minerva.pipeline	Utilities for training and evaluating models in a reproducible manner.
minerva.samplers	Custom data samplers to enable advanced sampling strategies during data loading.
minerva.transforms	Data transformations and augmentations.
minerva.utils	General utility functions and helper classes.

Supervised Learning

For simply using Minerva to train a model, you may check the following notebooks that provide a step-by-step guide on how to use Minerva for different tasks:

• Getting Started with Minerva for Seismic Facies Classification

• Getting Started with Minerva for Human Activity Recognition

Minerva Design

Minerva is designed to be modular and extensible, allowing users to integrate and extend its functionalities with ease. Checkout the Minerva design to understand how Minerva is structured and how to extend it.

Also, minerva provides a powerful experimental framework that allows you to run experiments in a reproducible manner. The experimental framework is built on top of PyTorch Lightning and provides a set of tools to manage experiments, organize and log results, and evaluate model. You may refer to the Minerva experimental framework to understand how to use it.

Self-Supervised Learning

Under construction