ZADU¶
ZADU measures what changes when high-dimensional data is projected into fewer dimensions. Its 22 measures cover neighborhoods, distances, class structure, density, and empty regions in scatterplots.
Provide the original data and a projection of the same samples. ZADU returns one result for each measure you select.
from sklearn.datasets import load_iris
from sklearn.decomposition import PCA
from zadu import ZADU
original, labels = load_iris(return_X_y=True)
projection = PCA(n_components=2).fit_transform(original)
scores = ZADU([{"id": "tnc", "params": {"k": 10}}], original).measure(projection)
print(scores[0]) # trustworthiness and continuity; higher is better
Find what you need¶
| Task | Documentation |
|---|---|
| Decide what to measure and interpret the result | Choose measures |
| Look up parameters, return keys, and papers | Measure reference |
| Compare projections of the same dataset | Evaluate many projections |
| Locate distortion in a scatterplot | Pointwise scores and Visualization |
| Control memory use | Memory and execution |
| Run supported calculations with MLX or PyTorch | Execution backends |
| Add a measure or report a problem | Contributing |
Citation¶
Cite the ZADU paper and the software version you used. The measure reference links each measure's original paper.