Free NVIDIA Accelerated Data Science certification exam Practice Questions

Sample Practice Questions

Try these 10 sample questions from the NVIDIA Accelerated Data Science certification exam practice bank. Each includes the answer and explanation. Continue with the interactive practice above to review the full question set.

1. 1A traffic level is typical during the day but unusual when the office is closed at night. Which anomaly type describes the nighttime observation?

   1. A.A contextual anomalyCorrect answer: A
   2. B.A point anomaly independent of context
   3. C.A collective anomaly of a sequence
   4. D.A change point between two regimes

   Explanation: The source says "Contextual anomalies refer to data points that deviate significantly within a specific context but appear normal outside of that context." The time of day provides the context for this unusual observation.

   Reference: Anomaly Detection for Time Series Data: Anomaly Types — Contextual Anomalies

2. 2A Dask cuDF ETL pipeline needs a recommended columnar file format that enables column projection and predicate pushdown. Which format does the Dask cuDF guide recommend?

   1. A.ParquetCorrect answer: A
   2. B.CSV
   3. C.JSON Lines
   4. D.Plain-text log

   Explanation: The Dask cuDF guide states: "Parquet is the recommended file format for Dask cuDF. It provides efficient columnar storage and enables Dask to perform valuable query optimizations like column projection and predicate pushdown." Parquet provides the stated columnar optimization.

   Reference: Dask cuDF Best Practices — Reading Data: Use Parquet

3. 3A cuDF DataFrame should identify duplicate rows using only customer_id, even when other columns differ. Which argument controls the comparison columns?

   1. A.subset=['customer_id']Correct answer: A
   2. B.keep='last'
   3. C.inplace=True
   4. D.ignore_index=True

   Explanation: The API says of subset: "Only consider certain columns for identifying duplicates, by default use all of the columns." Restricting subset to customer_id compares that column.

   Reference: cudf.DataFrame.drop_duplicates — NVIDIA cuDF — Parameters

4. 4A cuGraph edge list omits isolated vertices. When providing the vertices argument to from_cudf_edgelist, which IDs must the argument contain?

   1. A.All vertices, including those already in the edge listCorrect answer: A
   2. B.Only isolated vertices absent from the edge list
   3. C.Only vertices that appear as edge sources
   4. D.Only vertices that appear as edge destinations

   Explanation: The cuGraph API says: "If specified, this array must contain every vertex identifier, including vertex identifiers that are already included in the source and destination arrays." The list includes isolated and edge-connected vertices.

   Reference: cugraph.Graph.from_cudf_edgelist — NVIDIA cuGraph — Parameters: vertices

5. 5A classification dataset has imbalanced classes, and each cross-validation fold should approximately retain the overall class proportions. Which splitter directly addresses that goal?

   1. A.StratifiedKFoldCorrect answer: A
   2. B.KFold without stratification
   3. C.LeaveOneOut
   4. D.TimeSeriesSplit

   Explanation: The scikit-learn guide states: "StratifiedKFold is a variation of K-fold which returns stratified folds: each set contains approximately the same percentage of samples of each target class as the complete set." It approximately preserves class proportions in each fold.

   Reference: Cross-validation: evaluating estimator performance — scikit-learn — 3.1.2.2.1 Stratified K-fold

6. 6A cuDF object-dtype feature has many repeated labels. Which representation does the cuDF guide suggest for efficient storage?

   1. A.Categorical codes with a shared category dictionaryCorrect answer: A
   2. B.A full string value stored for each row
   3. C.A separate one-hot column for each label
   4. D.A separate label dictionary stored for each row

   Explanation: The cuDF guide recommends: "Use a Categorical for efficient storage of an object-dtype column with many repeated values." Its categorical accessor lists: "categories The categories of this categorical. ordered Whether the categories have an ordered relationship. codes Return Series of codes as well as the index." Categories and codes support this representation.

   Reference: cudf.DataFrame.memory_usage — NVIDIA cuDF — Examples

7. 7A recurring sequence of individually ordinary measurements forms an unusual pattern when considered together. What type of anomaly does the sequence illustrate?

   1. A.An isolated value outside the overall distribution
   2. B.A collective anomaly in the sequenceCorrect answer: B
   3. C.A single observation unusual only for its time of day
   4. D.A single value unusually distant from its local trend

   Explanation: The reference states: "Collective anomalies involve a set of data points that, when considered together, deviate from what is expected, even though individual points may not qualify as anomalies." The unusual sequence is therefore a collective anomaly.

   Reference: Anomaly Detection for Time Series Data: Anomaly Types — Collective Anomalies

8. 8A Dask cuDF ETL pipeline uses map_partitions for a custom transformation. How can the pipeline mitigate the loss of column-projection pushdown around this opaque expression?

   1. A.Select needed columns before and after map_partitionsCorrect answer: A
   2. B.Select all columns after every map_partitions call
   3. C.Compute the full input on the client before filtering
   4. D.Repeat the custom transformation on every column

   Explanation: The guide says: "Since column-projection pushdown is often the most effective optimization, it is important to select the necessary columns both before and after calling dask.dataframe.DataFrame.map_partitions()." Explicit selection limits columns around the opaque transformation.

   Reference: Dask cuDF Best Practices — User-defined functions

9. 9For duplicate rows, what does cuDF drop_duplicates(keep=False) do?

   1. A.It removes all occurrences of each duplicated rowCorrect answer: A
   2. B.It retains only the first occurrence of each duplicate
   3. C.It retains only the last occurrence of each duplicate
   4. D.It retains every occurrence of each duplicate

   Explanation: The cuDF API specifies: "- ‘first’ : Drop duplicates except for the first occurrence. - ‘last’ : Drop duplicates except for the last occurrence. - False : Drop all duplicates." False removes all duplicate occurrences.

   Reference: cudf.DataFrame.drop_duplicates — NVIDIA cuDF — Parameters

10. 10A cuGraph analyst needs source-to-vertex shortest paths on an unweighted graph. Which algorithm does the sssp documentation direct the analyst to use?

    1. A.Breadth-first searchCorrect answer: A
    2. B.Weighted single-source shortest paths
    3. C.PageRank
    4. D.Connected components

    Explanation: The sssp reference states: "Unweighted graphs are also unsupported. For finding shortest paths on an unweighted graph, use BFS instead." BFS is the documented choice for the unweighted case.

    Reference: cugraph.sssp — NVIDIA cuGraph — cugraph.sssp description