Advanced Analytics
Free NVIDIA-Certified Professional: Accelerated Data Science practice — 6 questions on Advanced Analytics, with explanations. No sign-up. Full 12-question mixed test →
A financial services company runs fraud detection on a transaction graph with 50M nodes and 200M edges on a DGX cluster. Analysts need to surface tightly-connected clusters of accounts that transact heavily with each other, indicative of collusive fraud rings, rather than ranking individual account importance. Which cuGraph algorithm BEST accomplishes this?
Louvain is a community detection algorithm that partitions the graph to maximize modularity, directly surfacing tightly-connected clusters such as fraud rings.
In a social network graph, an analyst wants to identify accounts that act as bridges connecting otherwise separate communities, since these bottleneck accounts are critical for controlling information flow between clusters. Which cuGraph centrality measure is specifically designed to score nodes based on the fraction of all shortest paths that pass through them?
Betweenness centrality explicitly measures how often a node lies on the shortest path between other node pairs, making it the correct choice for identifying bridge/bottleneck nodes between communities.
You must flag anomalous readings in a 500M-row GPU dataframe of sensor telemetry sampled every 1 minute, using a trailing 24-hour rolling z-score in cuDF for a real-time monitoring pipeline. The window must use only past and current data with no look-ahead bias. Which rolling configuration is correct?
At 1-minute sampling, 24 hours equals 1440 periods, and center=False produces a trailing (causal) window that only uses past and current values, avoiding look-ahead bias in real-time anomaly detection.
A geospatial team needs to determine, for millions of GPS pings, which delivery zone polygon each ping falls inside, using GPU acceleration. Which capability does cuSpatial provide to support this at an awareness level within the RAPIDS ecosystem?
cuSpatial provides GPU-accelerated geospatial primitives including point-in-polygon testing, spatial joins, distance calculations, and trajectory analysis, which is exactly what determining polygon membership for GPS pings requires.
While engineering features for a 5M-row tabular dataset in cuDF, a data scientist computes a target-mean encoding for a high-cardinality categorical column by running a single groupby on the target across the entire dataset and merging the result back onto every row before training. Model validation accuracy is suspiciously high. Which approach correctly addresses the underlying problem?
Computing the target-mean encoding out-of-fold (excluding each row's own fold) prevents each row's own target value from leaking into its own feature, which is the root cause of inflated validation accuracy.
A recommendation engine needs influence scores that are biased toward relevance to a specific seed set of users who purchased Product X, rather than a global importance ranking across the entire graph. Which cuGraph approach is designed for this personalized use case?
cugraph.pagerank accepts a personalization parameter that biases the random-walk restart distribution toward specified seed nodes, producing personalized (topic-sensitive) PageRank scores relevant to the seed set.
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.
$129.99 $34.99 with code FREETEST33 — valid through September 9.
Get my $34.99 deal →