DART is a python package which facilitates distributional similarity measurements at both population and subpopulation levels.
Note
The detailed documentation for this repository can be accessed at didsr.github.io/DART/
DART encodes attribute (metadata) information into hypervectors (high-dimensional vectors) which can then be used combined to represent individual samples or entire datasets following the principles of hyperdimensional computing.
Hyperdimensional computing allows for the nuanced representation of different attribute types through the creation of hypervectors with varying amounts of similarity to each other. This enables proper representation of different types of attributes, such as drawing a distinction between categorical attributes (for which distinct values have no inherent similarity) and numeric attributes (for which the inherent similarity between any two values depends on their difference).
- clone this repository
- install dependencies (run from the root directory of this project):
- If you would like to run
test.ipynb:pip install ".[test]" - If you just want to install the DART package:
pip install "."
A full implementation example can be found in the test notebook.
- A. Burgon, N. Petrick, D. Krainal, A. Khan, and R. K. Samala, "Data Representativeness with Hyperdimensional Computing" Under review
- A. Burgon, N. Petrick, D. Krainal, A. Khan, and R. K. Samala, "Metadata distribution evaluation through hyperdimensional encoding", Accepted at SPIE Medical Imaging, 2026
RST Reference Number: TBD
Date of Publication: TBD
Recommended Citation: TBD