NVIDIA-Certified-Professional Accelerated Data Science Sample Questions:

1. A data engineer is preparing a dataset for training a deep learning model. The dataset contains numerical features with missing values, outliers, and inconsistent units.
Which of the following strategies is the most appropriate for ensuring a standardized and clean dataset?

A) Remove all rows with missing values and outliers to ensure only clean data is used.
B) Standardize the dataset using the mean and standard deviation, but keep missing values and outliers unchanged to avoid data manipulation.
C) Use the median to fill missing values, convert all numerical values into categorical bins, and apply Min-Max scaling.
D) Replace missing values with the mean, apply z-score normalization, and clip extreme outliers based on a threshold (e.g., 3 standard deviations).

2. A company is processing large log files from a cloud application, accumulating over 5TB of data daily. The data processing pipeline must be GPU-accelerated to extract insights quickly.
Which of the following is the most effective approach to handle high-volume log processing using NVIDIA technologies?

A) Use cuDF with explicit memory management to load and process the entire dataset into a single GPU.
B) Store logs as Pandas DataFrames and use multiprocessing to parallelize operations across CPU cores.
C) Leverage Dask-cuDF to distribute the dataset across multiple GPUs, ensuring efficient parallel processing.
D) Use RAPIDS cuML for performing log file processing, taking advantage of its optimized ML algorithms.

3. You are working with structured tabular data in a cloud-based GPU environment.
Your dataset contains the following columns:
Column Name Example Values Data Type Needed
user_id 15432, 98765, 43210 Integer
purchase_amt 12.99, 35.50, 100.75 Floating Point
category 'Books', 'Electronics' Categorical
Which of the following is the most optimal approach to assign data types to these columns to ensure efficient memory usage and computational performance?

A) 1. df['user_id'] = df['user_id'].astype('int16')
2. df['purchase_amt'] = df['purchase_amt'].astype('float16')
3. df['category'] = df['category'].astype('string')
B) 1. df['user_id'] = df['user_id'].astype('int64')
2. df['purchase_amt'] = df['purchase_amt'].astype('float64')
3. df['category'] = df['category'].astype('string')
C) 1. df['user_id'] = df['user_id'].astype('float32')
2. df['purchase_amt'] = df['purchase_amt'].astype('float64')
3. df['category'] = df['category'].astype('string')
D) 1. df['user_id'] = df['user_id'].astype('int32')
2. df['purchase_amt'] = df['purchase_amt'].astype('float32')
3. df['category'] = df['category'].astype('category')

4. You need to deploy a machine learning model on a GPU-equipped system. The GPU has 16GB of VRAM, and the model requires approximately 12GB of memory during inference. However, additional system processes and other applications consume 5GB of VRAM.
What would happen if you attempt to run inference without making any optimizations, and how should you resolve the issue?

A) The model will run without issues because 16GB of VRAM is sufficient for a 12GB model
B) Switching from a GPU to CPU inference will resolve memory issues without performance loss
C) The model will fail to run due to out-of-memory (OOM) errors, and using a smaller batch size can help reduce memory usage
D) The model will run successfully but with reduced performance due to memory fragmentation

5. A data scientist is working with large-scale datasets in a RAPIDS AI pipeline and needs to efficiently process and organize the data while leveraging GPU acceleration.
Which of the following approaches best ensures optimized processing and memory management when using NVIDIA technologies?

A) Use cuDF to store and process data in GPU memory, leveraging its vectorized operations for transformations and aggregations.
B) Store data in a pandas DataFrame first and then convert it to cuDF only when GPU operations are needed.
C) Write intermediate data to disk as CSV files before reading them back into RAPIDS AI to avoid memory overflow.
D) Process data using Apache Spark on CPU before transferring results to GPU for model training.

Solutions: