Advanced and Graph Analytics
Free NVIDIA-Certified Associate: Accelerated Data Science practice — 6 questions on Advanced and Graph Analytics, with explanations. No sign-up. Full 12-question mixed test →
A GPU data science team uses cuML to analyze 50 million credit-card transactions for potential fraud. They want an unsupervised technique that flags transactions in low-density regions of feature space as suspicious, without requiring a pre-labeled fraud dataset or specifying the number of anomaly clusters in advance. Which cuML approach BEST fits this requirement?
DBSCAN is density-based, requires no predefined cluster count, and naturally labels points in low-density regions as noise (-1), which directly serve as anomaly/fraud candidates — matching every stated requirement.
A citation network analysis needs to determine paper importance based on the structure of citation links, where a paper is considered more important if it is cited by other important papers (recursive, link-based importance). Which cuGraph algorithm best fits this requirement?
PageRank explicitly computes node importance recursively, weighting incoming links by the importance of the source nodes — the exact definition of the scenario.
A network resilience scenario requires identifying the small set of nodes whose removal would most likely fragment the network, because these nodes lie on the largest number of shortest paths between other node pairs. Which cuGraph centrality algorithm should be used?
Betweenness centrality quantifies how often a node lies on shortest paths between other pairs — high values identify critical bridge/bottleneck nodes whose removal fragments the graph.
You build a cuGraph Graph from an edge list containing large, sparse, non-contiguous vertex identifiers (e.g., customer IDs like 8834471 and 92110345), calling G.from_cudf_edgelist() with default settings. What actually happens, and why does this matter for the exam?
By default renumber=True in cuGraph's Graph construction, automatically mapping arbitrary/sparse vertex IDs into a compact contiguous internal index range, which dramatically improves GPU memory usage and algorithm performance — a key internal behavior the exam tests.
A social network analysis task requires community detection where every returned community must be internally well-connected, with no disconnected or poorly-connected subsets under a single community label — a known limitation observed with modularity-based Louvain on large graphs. Which cuGraph algorithm should be selected instead?
cuGraph's Leiden implementation directly addresses Louvain's documented weakness of sometimes producing internally disconnected communities, guaranteeing well-connected community output.
In cuGraph, what is the fundamental difference between calling cugraph.bfs() versus cugraph.sssp() on the same graph object to compute distances from a source vertex?
BFS traverses level-by-level, treating every edge as weight 1, and is intended for unweighted graphs. SSSP (Dijkstra-based) uses actual edge weight values to compute minimum-weight paths on weighted graphs — this is the core distinction tested on the exam.
Ready for the real thing?
The full course: two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed explanations.
$99.99 $34.99 with code FREETEST33 — valid through September 7.
Get my $34.99 deal →