最新的NVIDIA NCP-ADS免費考試真題

問題1

A machine learning engineer is working with a financial dataset that contains multiple numerical features, including income, loan amount, and transaction frequency. Some features are normally distributed, while others have a highly skewed distribution with extreme outliers.
Which of the following approaches best ensures uniformity across features before training a model?

A. Remove outliers before applying standardization

B. Scale all numerical features using min-max normalization

C. Apply log transformation to skewed features before standardizing them with z-score normalization

D. Use one-hot encoding to transform numerical features into categorical representations

正確答案: C

問題2

A data scientist is analyzing a large time-series dataset containing stock price movements of thousands of companies over a decade. The dataset is stored as a cuDF DataFrame and contains millions of rows. The scientist wants to visualize trends and patterns interactively while leveraging GPU acceleration.
Which of the following approaches is the most efficient for visualizing this time-series data?

A. Use matplotlib with plt.plot() while applying df.to_pandas() to convert data.

B. Use cuXfilter with a cuDF DataFrame to generate interactive visualizations directly on the GPU.

C. Convert the cuDF DataFrame to Pandas and use matplotlib for plotting.

D. Use seaborn with a sampled subset of the dataset to generate line plots.

正確答案: B

問題3

A data science team wants to deploy a GPU-accelerated pipeline using cuGraph to analyze graph data on cloud infrastructure. They are evaluating different cloud-based GPU solutions.
Which of the following factors should they consider when selecting a cloud-based GPU instance for running cuGraph efficiently?

A. cuGraph runs equally well on CPU-based virtual machines, making GPU instances unnecessary.

B. Cloud-based GPUs are only useful for rendering graphics, not for running cuGraph algorithms.

C. The availability of NVIDIA CUDA-enabled GPUs, as cuGraph requires CUDA for acceleration.

D. The choice of GPU instance does not affect cuGraph performance since all GPUs execute graph algorithms at the same speed.

正確答案: C

問題4

In the context of cloud computing, what are the key benefits of using GPUs for data science tasks?
(Select two)

A. Better for memory-intensive workloads

B. Lower cost of cloud infrastructure

C. Faster parallel processing for large datasets

D. Lower energy consumption compared to CPUs

E. Efficient handling of matrix operations in machine learning models

正確答案: C,E

問題5

You are training a machine learning model using RAPIDS cuML and need to ensure that all numeric features are standardized for better model performance.
Which of the following is the best approach for scaling data using RAPIDS?

A. df_scaled = df / df.max()

B. df_scaled = (df - df.min()) / (df.max() - df.min())

C. df_scaled = scaler.fit_transform(df)

D. scaler = cuml.preprocessing.StandardScaler()

E. df_scaled = df.apply(lambda x: x / np.linalg.norm(x))

正確答案: D

問題6

A retail company is deploying an AI-driven demand forecasting system using NVIDIA GPUs. The team follows the CRISP-DM framework and is currently in the Evaluation phase.
Which approach best leverages NVIDIA technologies to assess model performance effectively?

A. Use RAPIDS cuML to rapidly compute evaluation metrics like RMSE and R-squared on large datasets using GPUs.

B. Assume that a high training accuracy guarantees excellent real-world performance, skipping the evaluation phase.

C. Rely only on training loss as the primary evaluation metric without considering validation performance.

D. Perform evaluation on a small CPU-based subset of the dataset instead of using full GPU-accelerated inference.

正確答案: A

問題7

Which feature of NVIDIA MLFlow integration with Triton Inference Server allows for the seamless deployment and monitoring of models in production?

A. MLFlow handles only the training and not the deployment of models.

B. MLFlow prevents models from being updated once deployed to production.

C. MLFlow allows for version control and automated model rollout.

D. MLFlow only supports deployment on CPUs.

正確答案: C

問題8

A machine learning engineer is tasked with optimizing an image classification model on a cloud platform. The engineer must select a GPU-accelerated instance that balances cost and performance while ensuring compatibility with frameworks like TensorFlow and PyTorch.
Which instance configuration is the most appropriate choice?

A. A single-core CPU instance with high disk I/O throughput for faster data loading.

B. A cloud instance with NVIDIA A100 GPUs and NVLink support.

C. A cloud instance with integrated graphics rather than dedicated NVIDIA GPUs.

D. A CPU-only instance with 128 GB of RAM for increased data processing speed.

正確答案: B

問題9

You are processing a large dataset in a distributed computing environment using RAPIDS and Dask.
Your workflow involves frequent shuffling of data between partitions, leading to significant slowdowns.
Which of the following strategies is the best way to implement data caching to reduce shuffle overhead using NVIDIA technologies?

A. Enable GPU-accelerated caching with RAPIDS cuDF and persist intermediate results in GPU memory.

B. Use traditional disk-based caching by writing intermediate results to CSV files and reloading when needed.

C. Disable caching altogether to force a recomputation of results, ensuring up-to-date data processing.

D. Use a CPU-based caching solution like Memcached to store intermediate data before reloading into cuDF.

正確答案: A

問題10

You are processing a large dataset using NVIDIA Dask-cuDF to distribute GPU-accelerated computation across multiple nodes. Users report inconsistent execution times, with some jobs taking significantly longer than expected.
Which of the following actions would best help diagnose the performance bottleneck?

A. Use Dask's dashboard and NVTX markers to analyze task execution times and GPU utilization.

B. Limit GPU memory usage to force more frequent spilling to disk and observe performance differences.

C. Reduce the number of Dask workers to minimize parallel execution overhead.

D. Switch to using Pandas with Dask to compare execution speed differences.

正確答案: A

最新的NVIDIA-Certified-Professional Accelerated Data Science - NCP-ADS免費考試真題

聯系我們

站內鏈接

最新更新