Network Machine Learning Project - Drosophila brain connectome analysis

Installation

conda install cuda-toolkit=11.7 -c nvidia
pip install -r requirements.txt -f https://data.pyg.org/whl/torch-2.0.0+cu117.html

Note that the URL above needs to be changed depending on the PyTorch and CUDA Toolkit versions as well as the OS. Find the correct URL on the Installation page of PyG's documentation.

Reproducibility

1. Download the raw data from FlyWire Codex. We used the April 2023 version of the dataset (Snapshot 630).
2. Run scripts/construct_graph.py to generate the NetworkX DiGraph object based on the raw data. The result is saved as a pickle file.
3. Run scripts/build_node_embedding.py to generate the Node2Vec and spectral node features. The results are Pandas DataFrames saved as pickle files. This might take several hours.
4. Follow notebooks/exploration.ipynb to reproduce figures in the Exploration section of the report.
5. Follow notebooks/exploitation.ipynb to train the models discussed in the Exploitation section of the report. The accuracies and F1 scores of the models in different task configurations are saved in a CSV file. Note that it takes about an hour to train all models with a GPU.
6. Follow notebooks/model_comparison.ipynb to reproduce the result benchmarking figure and the confusion matrices.