Auto-Skong – short for Auto-Scikit-Longitudinal – is an Automated Machine Learning (AutoML) library for Longitudinal Data Analysis, with a focus on Longitudinal ML Classification tasks!
Get Started 🚀
Paper is incoming. Submitted to a conference.
What's Auto-Sklong? ¶
A brand-new Search Space tailored for Longitudinal ML Classification tasks
Built upon GAMA, Auto-Sklong
offers today a brand-new search space, tailored for Longitudinal Machine Learning Classification tasks.
Such search space combines both longitudinal and non-longitudinal machine learning techniques,
making it the first in the literature to the best of our knowledge. By leveraging the Scikit-Longitudinal,
our newly-developed machine learning library for longitudinal machine learning classification tasks,
and Scikit-Learn, we are able to provide a unique search space
for the Combined Algorithm Selection and Hyperparameter Optimisation (CASH) problem in the context of Longitudinal Machine Learning Classification tasks.
For further information, please refer to our Search Space page.
To search through the search space proposed, Auto-Sklong comes with a variety of search methods
Bayesian Optimisation via SMAC3,
Random Search, Successive Halving, and Evolutionary Algorithms via GAMA.
Your data's temporal dependency matter
While designing Scikit-Longitudinal,
which in a nutshell is a machine learning toolkit for longitudinal data classification tasks, complying with the Scikit-Learn API,
We came up with a very easy and simple
way to understand the temporal dependencies of your data. We
rely on a (user-)defined matrix representation of which feature indices are
both temporal and or non-temporal. Which hence for temporal
features how recent they are. Therefore, by conveying to this
matrix, we let algorithms' designers, within with the ability to
deal with the temporal dependencies of the data the way they
designed their approach to cope with longitudinal data classification
using machine learning. Read more about it in our Temporal dependency page.
Now, given that the introduced new search space contains longitudinal machine learning techniques from
Scikit-Longitudinal, we
use the same matrix representation to convey the temporal dependencies of the data to the algorithms
throughout the AutoML system's search.
Your Longitudinal Analysis using AutoML in ~10 lines of code
Auto-Sklong aims to be simple to use and understand.
By following the popular Fit/Predict/Transform API style of Scikit-learn, we allow any
user already familiar with the library to take up to a couple of seconds to use the library,
and for newcomers a couple of minutes, as Scikit-learn is the, state-of-the-art
in the ease of use and standard machine learning algorithms in the community today.
Anyway, here is a simple example of how to use Auto-Sklong to perform an
AutoML search on your Longitudinal Machine Learning Classification task.
Open Source Software at the core of our team
Auto-Sklong is an open source project. We believe in open-source software and the community that supports it. We are dedicated to contributing to the open-source community and working with other projects.
Who's our team? ¶
Simon Provost
Main Author of Auto-Sklong & Ph.D Student
@ University of Kent
Simon designed, developed and maintain the Auto-Sklong library. Simon is a Ph.D student in computer science at the University of Kent, United-Kingdom. His research interests are in Machine Learning (ML), Automated Machine Learning and ML-applied Healthcare.
Prof. Alex Freitas
Professor of Computational Intelligence, Head of Research
@ University of Kent
Alex is a Professor of Computational Intelligence at the University of Kent, United-Kingdom. He is the main supervisor of Simon's Ph.D., and also designed the Auto-Sklong library with. Alex's research interests are in machine learning, data mining, and bioinformatics.