Data Analysis and Visualization

NCA-GENL · 42 questions

• A city parks desk has a 180-row picnic-table repair log that already opens in a desktop spreadsheet. A vendor says RAPIDS is required before anyone can chart repair hours versus lumber cost. When is GPU-accelerated table work justified?
• A county clerk wants one analysis pass over marriage-license filings, courtroom calendars, and parking-ticket ledgers that together will not fit in a laptop session. Which official NVIDIA stack loads, joins, and summarizes those tables on GPU?
• A ferry-terminal authority has a decade of boarding tickets and a decade of tide-gauge readings. Staff export slices to a workstation and the joins time out. How should they integrate the two ledgers?
• A public-works yard filters pothole reports on the GPU, copies the result back to host memory to be safe, then sends it to the GPU again for a group-by. What is the problem with that bounce?
• A city-archives desk wants page-count spread, missing box labels, and how many lots came from each branch before anyone fine-tunes a model. A junior treats those plots as the experiment. What are the plots?
• A community-garden inventory exports a wide parquet of daily soil moisture and plot IDs. The analyst is about to open it in pandas and then move it to the GPU. How should that table land on a single GPU?
• A river-gauge station's 15-year stage archive is larger than the memory of the one GPU on the bench. How should they ingest it without discarding the science?
• A landfill scale-house now has two GPUs on the same box and a truck-ticket archive that barely fit on one device last season. How should they match the reader to scale?
• A school-bus depot is handed ridership files, delay logs, and weather extracts that use different date columns. Staff want to concatenate first and clean later. What is the right ingest order?
• A public-library intern writes a Python loop that walks every checkout row to recode a material label. The table is columnar and already on the GPU. What should they use instead?
• A municipal animal-shelter log has mixed string and number columns, blank feeding times, and volunteer rows that should be dropped. How should that prep stay on the GPU?
• A city streetlight shop wants hour-of-day, day-of-week, and a one-hot fixture code from a GPU table of work orders before any estimator runs. How should those columns be built?
• A planning-department intern asks why cuDF tables can be handed to other RAPIDS steps without a CSV round-trip. What keeps those GPU frames cheap to share?
• A city-hall records desk is building an LLM help corpus from ordinance PDFs. Before anyone trains, they need a column of token counts per page so they can drop empty pages and flag oversized ones. What is token length in that step?
• A wastewater-plant desk wants cuML clustering on temperature, turbidity, and chlorine. Raw units differ by orders of magnitude. What must happen before that GPU estimator runs?
• A bridge-inspection office shuffles report IDs into train and eval files but copies any row that mentions urgent into both so the eval looks realistic. What defect is that?
• A municipal pool desk wants a classifier for too-cold, too-hot, and just-right water notes. Almost every labeled slip says just right. What should they do first?
• A county-museum chatbot corpus has a pile of one-word captions and a few entire catalog PDFs pasted as one document. What plot helps them cap, split, or drop before ingest?
• A farmers-market inspector's scorecards omit the stall name on rainy Saturdays. A missingness plot shows that pattern. How should the desk read it?
• A community-theater costume shop has plenty of costume-ripped notes and almost no wig-melted notes. Staff propose paraphrasing the rare notes with a human check. What is that step?
• A municipal tree-nursery wants other desks to reuse their labeled planting notes. What short document records who collected the rows, the license, what the labels mean, and what the set is not for?
• A courthouse FAQ set was split by filename. Many eval questions are the train questions with a comma moved, and scores look perfect. What is the issue?
• A lighthouse-archive scrape keeps navigation chrome, click-here lines, and empty pages. What preprocessing step comes before any length plot or index build?
• A city recycling-sort yard already has a cleaned cuDF table of bin weights and contamination flags. Staff want scikit-learn-style fit and predict without leaving the GPU. Which official library is that?
• A civic-auditorium lost-and-found encodes item notes and wants to look at groups before anyone labels them. How should they explore that structure on the GPU?
• A municipal irrigation desk has labeled dry, seeping, and flooded sensor rows already on the GPU. How should they apply a supervised GPU estimator?
• A coastal-park ranger desk has a multi-GPU box and a table of trail-use, rainfall, and campsite-occupancy features that no longer fit on one device. How should they train boosted trees at that scale?
• A city marching-band warehouse intern offers to write a CUDA kernel to train boosted trees on the GPU table. What should they use instead?
• After a cuML cluster step, one cluster is almost all garbled OCR lines from a city tax-assessor's scanned invoices. How should that cluster be used?
• A city scooter-share desk has a huge two-column file of from-station and to-station trips. How do they get that edge table onto the GPU for analysis?
• A city parade-route planner's visitor-path graph has tens of millions of edges. NetworkX on a laptop will not finish overnight. Where do massive-scale graph analytics belong?
• After a cuGraph importance pass, a city food-bank warehouse wants those scores as extra columns on the original product table for later exploratory analysis. What should they do?
• A civic-arboretum FAQ is being loaded into a vector store. One pass cuts every 2,000 characters with no overlap, and questions that straddle a cut come back empty. Which ingest knobs matter?
• After encoding trail-marker plaques at a city nature preserve, a 2D sketch of the vectors shows hours and closing time in one clump and OCR garbage in another. What is that sketch?
• A city water-plant safety index was built with encoder A. A later desk embeds new PDFs with encoder B and appends them, and neighbors become nonsense. What must ingest record?
• A civic-orchestra handbook has numbered sections, but an ingest job still slices every N tokens, often in mid-sentence. How should those chunks be cut?
• A city aquarium desk is told to pick any embedder off the internet for a new visitor-FAQ index. Which official NVIDIA path is the associate retrieval and embedding stack?
• A county-fair transcript dump contains identical copies, copies with a typo, and paraphrased repeats of the same safety briefing. How does NeMo Data Curator distinguish those at selection depth?
• After dedup, a county-fair dump still has two-word fragments, other-language ads, and highly repetitive spam. What should happen before anyone builds a vector index?
• A rec-center desk's chemical, hours, and class-signup labels disagree on the same sentence, and named-entity spans for locker numbers sometimes include the next word. What data work comes first?
• A municipal print-shop wants author attribution on press notes and a small extractive question-answering set over equipment manuals. One author's notes are three words; another writes pages. What should they plot first?
• A lock-and-dam inspection office needs load, split, embed, and write-to-store as a repeatable pipeline, with a branch when a file fails to parse. Which official tools orchestrate that as data work?