A neural architecture search strategy proposed in "Convolutional networks for video-based infant movement analysis". This repository comprises the Python template and pseduocode of K-Best Search.
The Python template of K-Best Search is available in kbestsearch.py.
To adapt K-Best Search to your task, make the following two modifications to kbestsearch.py:
- Define the choices and associated alternatives you would like the search space to consist of by modifying the search_space dictionary at line 202.
- Include in the trainval method (line 209) all logic required to harness the selected alternatives in candidate to build a machine learning model, load data relevant for your task, perform training of the model, and estimate the performance of the model on a validation set. The estimated performance should consequently be returned by the method.
After adjusting the search space and trainval method in accordance with your needs, you are good to go:
python kbestsearch.py
The research is performed as a collaboration between Department of Neuromedicine and Movement Science and Department of Computer Science at the Norwegian University of Science and Technology and Norwegian Open AI Lab.
In case you found this work helpful in your research, we would appreciate that you cite the following:
@article{groos2022convolutional,
title={Convolutional networks for video-based infant movement analysis. Towards objective prognosis of cerebral palsy from infant spontaneous movements},
author={Groos, Daniel},
year={2022},
publisher={NTNU}
}
- K-Best Search for prediction of cerebral palsy from infant spontaneous movements with graph convolutional networks: "Development and Validation of a Deep Learning Method to Predict Cerebral Palsy From Spontaneous Movements in Infants at High Risk" (Groos et al., 2022) and "Fully automated clinical movement analysis from videos using skeleton-based deep learning" (Groos et al., 2021).