We're so confident of our products that we provide no hassle product exchange.
NVIDIA-Certified-Professional Accelerated Data Science Sample Questions:
1. A financial services company is using NVIDIA RAPIDS cuML to train a credit risk assessment model.
The dataset contains hundreds of numerical and categorical features, including loan amount, credit score, income, employment history, and previous loan defaults.
To optimize feature selection using NVIDIA technologies, which approach should they take?
A) Avoid feature selection, as modern deep learning models are capable of automatically handling redundant and irrelevant features.
B) Convert all numerical features into categorical variables to simplify the data and reduce the need for feature selection.
C) Use cuML's feature selection algorithms, such as Recursive Feature Elimination (RFE), to identify the most important predictors.
D) Keep all features in the dataset to ensure the model captures as much information as possible, even if some features are redundant.
2. A data scientist is training a deep learning model on an NVIDIA GPU but is encountering out-of- memory (OOM) errors.
To optimize GPU memory usage while maintaining efficient training performance, which of the following strategies should they prioritize?
A) Storing all training data in GPU memory at once
B) Increasing batch size without adjusting the optimizer settings
C) Using mixed precision training with automatic loss scaling
D) Using single-precision (FP32) calculations for better accuracy
3. Which of the following is the main advantage of using TensorRT for inference in an accelerated data science pipeline?
A) TensorRT automatically builds training models from raw data without requiring pre-trained models.
B) TensorRT optimizes deep learning models to run efficiently on NVIDIA GPUs by reducing precision while maintaining accuracy.
C) TensorRT is only compatible with image classification models and does not support other model types.
D) TensorRT is mainly used for data visualization and not for model inference.
4. You are working on a large-scale machine learning pipeline that involves processing massive datasets using multiple GPUs on an NVIDIA DGX system. You choose to use Dask to enable efficient parallel processing across multiple GPUs.
Which of the following steps is essential to correctly configure Dask for multi-GPU acceleration?
A) Avoid using dask_cudf and instead rely on standard pandas DataFrames to ensure GPU-accelerated execution.
B) Use dask_cuda.LocalCUDACluster() to create a cluster of GPU workers and pass it to the Dask client.
C) Use dask.distributed.Client() without specifying a scheduler to automatically detect available GPUs.
D) Assign computation tasks explicitly to CPUs using dask.config.set({'scheduler': 'threads'}) before using GPUs.
5. Which of the following methods are commonly used to handle missing data in data analysis? (Select two)
A) Using a fixed constant value for imputation
B) Removing rows with missing data
C) Ignoring missing data entirely
D) Imputing with the mean or median of the feature
E) Using the mode to impute missing values
Solutions:
| Question # 1 Answer: C | Question # 2 Answer: C | Question # 3 Answer: B | Question # 4 Answer: B | Question # 5 Answer: B,D |
By Julian
