Little Ball of Fur

Abstract

Sampling graphs is an important task in data mining. In this paper, we describe Little Ball of Fur a Python library that includes more than twenty graph sampling algorithms. Our goal is to make node, edge, and exploration-based network sampling techniques accessible to a large number of professionals, researchers, and students in a single streamlined framework. We created this framework with a focus on a coherent application public interface which has a convenient design, generic input data requirements, and reasonable baseline settings of algorithms. Here we overview these design foundations of the framework in detail with illustrative code snippets. We show the practical usability of the library by estimating various global statistics of social networks and web graphs. Experiments demonstrate that Little Ball of Fur can speed up node and whole graph embedding techniques considerably with mildly deteriorating the predictive value of distilled features.

Predictive Sampling Method for Spread Models in Networks

“…For a budget of n nodes in a sample, RN selects n random nodes from 𝑉, while RE selects enough edges from 𝐸 until the number of unique endpoints of edges (nodes) equals n (Ribeiro &Towsley, 2010). However, large networks seldom have all nodes and edges initially accessible or at least feasibly reachable (Rozemberczki et al, 2020b). A common reason is saving extremely large networks in relatively slow-acess storage mediums.Furthermore, organizations use systems with limited memory and therefore can only load and view a small proportion of a stored network.…”

“…The quality of sampled nodes can be evaluated by a metric that summarizes connectivity, clustering, degrees, or other characteristics of the sample. Many algorithms follow this general heuristic of starting at a single node 𝑠 and iteratively adding unvisited nodes 𝑣 in 𝑉 adjacent to nodes already in the sample (Rozemberczki et al, 2020b). To compare different exploration-based sampling algorithms, establishing some time or resource bound 𝐵 per sample creates an even field for comparisons.…”

“…Two problems current exploration-based network sampling algorithms face are addressed in this paper. First, targeting high-degree nodes in S was not deeply addressed during the development of many sampling algorithms (Rozemberczki et al, 2020b). In many real-world applications, node degree constitutes a value of importance (Rozemberczki et al, 2020b).…”

“…First, targeting high-degree nodes in S was not deeply addressed during the development of many sampling algorithms (Rozemberczki et al, 2020b). In many real-world applications, node degree constitutes a value of importance (Rozemberczki et al, 2020b). Actual meaning ranges from someone's social circle to the number of highways intersecting a city.…”

“…Targeting high-degree nodes provides direct benefit when sampling networks that model these applications. Second, a network sampling algorithm may linger in proximity to the start node 𝑠 rather than reach nodes farther away (Rozemberczki et al, 2020b). This renders 𝑆 a weaker representation of 𝐺 and forfeits opportunities to discover higher degree nodes farther away from 𝑠.…”