This repository provides HOTA implementation used for experiments in the papers.
Key Components:
-
src/method/hota.py: Implementation of the HOTA method (Section 4) -
notebooks: Interactive demonstrations of HOTA applied to datasets from the paper (Section 5)two_dimensional_benchmark.ipynb(Section 5.2)sphere_dataset_3_dim.ipynb(Section 5.3)
-
figures: Some experiments visualization logs -
requirements.txt: File with Python dependencies
├── README.md # Repository documentation
├── requirements.txt # Python dependencies
├── figures/ # Visualization produced notebooks from experiments directory
├── notebooks/ # Jupyter notebooks for paper experiments
│ ├── two_dimensional_benchmark.ipynb # Section 5.2
│ ├── sphere_dataset_3_dim.ipynb # Section 5.3 (3D case)
└── src/
├── datasets/
│ ├── datasets.py # Datasets w/o Potential barriers
│ └── lagrangian_potentials.py # Potential barriers
├── method/
│ ├── dynamics.py # Flow definition
│ ├── hota.py # Main HOTA algorithm
├── networks/
│ └── mlp.py # Time-augmented MLPs
└── utils/
└── metrics.py # Sinkhorn loss
-
Conda Environment Setup
# Create and activate conda environment conda create -n HOTA python=3.11 conda activate HOTA # Install dependencies pip install -r requirements.txt
-
Environment Configuration
# Base project directory export PROJECT_DIR=/absolute/path/to/HOTA # Python path configuration export SRC_DIR=${PROJECT_DIR}/src export PYTHONPATH="${SRC_DIR}:${PYTHONPATH}"
The notebooks below allow you to run HOTA on datasets from the paper and visualize results.
| Notebook | Paper Section | Comment |
|---|---|---|
| two_dimensional_benchmark.ipynb | 5.2 | Demonstrates HOTA on 2D benchmark tasks (e.g., Vneck, GMM) |
| sphere_dataset_3_dim.ipynb | 5.3 | Demonstrates HOTA on 3D Sphere dataset |