MegaBoost

Semi-supervised AutoML library on top of PyTorch.

About The Project

A state-of-the-art semi-supervised AutoML library on top of Pytorch to minimize your time and learning curve in machine learning. It is built for:

1. Semi-supervised model training with unlabeled data and very limited labeled data.

2. Fine tuning deep learning models.

3. Boosting model performance.

(back to top)

Installation

Make sure you have installed Pytorch and Torchvision from the official site. Then you can simply install this library from PyPI:

 pip install megaboost

Getting Started

Import the libraries:

 import torch
import megaboost as mg
labeled_dataset, unlabeled_dataset, test_dataset = mg.prepare_cifar10(resize=RESIZE)
megaboost = mg.MegaBoost(config=config)

Train the model using a similar style in scikit-learn:

 megaboost.fit(labeled_loader, test_loader, unlabeled_loader)

Use the model:

 res = megaboost.predict(image)

You can find the colab demo here.

MegaBoost Tutorial 1: Fine-tune Image Classification Model

Roadmap

• Enable MPS acceleration on Mac
• Enable automatic mixed precision by default
• SSL: image classification
• [ ]

(back to top)

Contact

maxnghello at gmail.com

Follow me on Medium: @medium.data.scientist

(back to top)

More

MegaBoost is an ensemble of state-of-the-art SSL methods with Self Meta Pseudo Labels as the backbone.