Welcome to the comparison-algorithms repository, where we explore and compare various embedding methods, specifically neuronal manifold techniques. Our novel algorithm, BunDLe-Net, is one of the focal points of this research. For the original implementation of BunDLe-Net, please visit its GitHub repository.
This repository is dedicated to evaluating and comparing different embedding methods as documented in the journal article on BunDLe-Net, accessible at https://www.biorxiv.org/content/10.1101/2023.08.08.551978v2.
Our repository is designed for easy replication of evaluations of embeddings for specific neuronal data and algorithms, eliminating the need to rerun redundant code.
- The core functions for BunDLe-Net can be found in the
functions/
directory. - Embeddings generated by various algorithms are produced by running Python scripts such as
1_PCA
,2_autoencoder
, and so on. The resulting embeddings are saved in thedata/generated/saved_Y/
directory. - Evaluations of these saved embeddings are performed using scripts located in the
evaluation_scripts/
directory. - Within
evaluation_scripts/
, you'll find scripts likemicrovariable_evaluation.py
,behaviour_decoding.py
, anddynamics_predictability.py
, which conduct various evaluations. - To run all evaluations for every worm and algorithm, you can use the
run_evaluations.sh
bash script. - Finally, all the plots of the evaluation metrics are done in plotting.py
In summary, if you wish to recompute an embedding using a specific method for a given dataset, consult the scripts labeled i_<algorithm>.py
, where 'i' ranges from 1 to 7. If you intend to re-evaluate a specific algorithm's embedding for a particular dataset, refer to the code within run_evaluations.sh
.