Datascape: exploring heterogeneous dataspace.

Area of Science:

• Data Science
• Computational Topology
• Graph Theory

Background:

• Current data science methods lack genericity and often overlook dataset shape.
• Understanding data structure and uncertainty is crucial for effective analysis.

Purpose of the Study:

• To introduce a novel framework, the datascape, for abstracting heterogeneous datasets.
• To leverage topology and graph theory to incorporate data shape into analysis.
• To enable exploration of the dataset's underlying space.

Main Methods:

• Utilized manifold learning and convex hull estimation principles.
• Constructed a framework combining nearest neighbor graphs, convex hulls, and shape-aware metric distances.
• Applied the datascape to simulated, ecological, and medical datasets.

Main Results:

• The datascape framework effectively uncovers underlying functions in simulated data.
• Predictive algorithms built with the datascape achieve performance comparable to state-of-the-art methods.
• Identified insightful geodesic paths within datasets, revealing underlying structures.

Conclusions:

• The datascape offers a versatile and powerful approach to data abstraction and analysis.
• Incorporating data shape enhances the understanding of complex datasets.
• The framework demonstrates broad applicability across various scientific domains.