Auto-Scikit-Longitudinal
(Auto-Sklong) is an Automated Machine Learning (AutoML) library, developed upon the
General Machine Learning Assistant (GAMA)
framework,
introducing a brand-new search space
leveraging both
Scikit-Longitudinal
and Scikit-learn
models to tackle the Longitudinal machine learning classification tasks.
Wait, what is Longitudinal Data — In layman's terms ?
Longitudinal data is a "time-lapse" snapshot of the same subject, entity, or group tracked over time-periods, similar to checking in on patients to see how they change. For instance, doctors may monitor a patient's blood pressure, weight, and cholesterol every year for a decade to identify health trends or risk factors. This data is more useful for predicting future results than a one-time survey because it captures evolution, patterns, and cause-effect throughout time.
Not enough?
- For more scientific details, you can refer to our paper published by
IEEE
in the IEEE International Conference on Bioinformatics and Biomedicine (BIBM), 2024 Edition. Auto-Sklong
comes with various search methods to explore thesearch space
introduced, such asBayesian Optimisation
. For more details, visit the official documentation.
Note
Want to use Jupyter Notebook
, Marimo
, Google Colab
, or JupyterLab
?
Head to the Getting Started
section of the documentation for full instructions! 🎉
To install Auto-Sklong:
-
✅ Install the latest version:
pip install auto-sklong
To install a specific version:
pip install auto-sklong==0.0.1
Caution
Auto-Sklong
is currently compatible with Python versions 3.9
only.
Ensure you have this version installed before proceeding.
This limitation stems from the Deep Forest
dependency.
Follow updates on this GitHub issue.
If you encounter errors, explore the installation
section in the Getting Started
of the documentation.
If issues persist, open a GitHub issue.
Here's how to run AutoML on longitudinal data with Auto-Sklong:
from sklearn.metrics import classification_report
from scikit_longitudinal.data_preparation import LongitudinalDataset
from gama.GamaLongitudinalClassifier import GamaLongitudinalClassifier
# Load your dataset (replace 'stroke.csv' with your actual dataset path)
dataset = LongitudinalDataset('./stroke.csv')
# Set up the target column and split the data (replace 'class_stroke_wave_4' with your target)
dataset.load_data_target_train_test_split(
target_column="class_stroke_wave_4",
)
# Set up feature groups (temporal dependencies)
# Use a pre-set for ELSA data or define manually (See docs for details)
dataset.setup_features_group(input_data="elsa")
# Initialise the AutoML system
automl = GamaLongitudinalClassifier(
features_group=dataset.feature_groups(),
non_longitudinal_features=dataset.non_longitudinal_features(),
feature_list_names=dataset.data.columns.tolist(),
max_total_time=3600 # Adjust time as needed (in seconds)
)
# Fit the AutoML system
automl.fit(dataset.X_train, dataset.y_train)
# Make predictions
y_pred = automl.predict(dataset.X_test)
# Print the classification report
print(classification_report(dataset.y_test, y_pred))
More detailed examples and tutorials can be found in the documentation!
If you use Auto-Sklong in your research, please cite our paper:
@INPROCEEDINGS{10821737,
author={Provost, Simon and Freitas, Alex A.},
booktitle={2024 IEEE International Conference on Bioinformatics and Biomedicine (BIBM)},
title={Auto-Sklong: A New AutoML System for Longitudinal Classification},
year={2024},
volume={},
number={},
pages={2021-2028},
keywords={Pipelines;Optimization;Predictive models;Classification algorithms;Conferences;Bioinformatics;Biomedical computing;Automated Machine Learning;AutoML;Longitudinal Classification;Scikit-Longitudinal;GAMA},
doi={10.1109/BIBM62325.2024.10821737}}
We enhanced @PGijsbers' open-source GAMA
initiative by introducing a brand-new search space designed specifically for tackling longitudinal classification problems. This search space is powered by our custom library, Scikit-Longitudinal
(Sklong), enabling Combined Algorithm Selection and Hyperparameter Optimization (CASH Optimization).
Unlike GAMA
or other existing AutoML libraries, Auto-Sklong
offers out-of-the-box support for
longitudinal classification tasks—a capability not previously available.
To better understand our proposed search space, refer to the visualisation below (read from left to right, each step being one new component to a final pipeline candidate configuration):
While GAMA
offers some configurability for search spaces, we improved its functionality to better suit our needs. You can find the details of our contributions in the following pull requests:
- ConfigSpace Technology Integration for Enhanced GAMA Configuration and Management 🥇
- Search Methods Enhancements to Avoid Duplicate Evaluated Pipelines 🥈
- SMAC3 Bayesian Optimisation Integration 🆕
Auto-Sklong is licensed under the MIT License.