NCP-ADS Study Guide

Domain Overview

Demonstrate understanding of exploratory data analysis, statistical foundations, and evaluation metrics.

Objectives

• Detect anomalies in time-series datasets.
• Conduct time-series analysis.
• Create and analyze graph data using GPU-accelerated tools such as cuGraph.
• Identify when dataset scale requires accelerated or distributed methods.
• Perform exploratory data analysis (EDA).
• Visualize time-series data effectively.

Objectives

• Design and implement accelerated ETL (extract, transform, load) workflows.
• Implement caching strategies to reduce shuffle overhead.
• Use distributed data processing frameworks for large-scale datasets.
• Implement Dask-based data parallelism for multi-GPU scaling.
• Profile deep learning workloads with tools such as DLProf.
• Choose optimal data processing libraries for varying dataset sizes and workloads.

Objectives

• Perform data cleansing and preprocessing with cuDF and pandas.
• Transform and standardize data for model readiness.
• Apply standardization to ensure feature uniformity where required.
• Generate synthetic data for augmentation using cuDF and RAPIDS.
• Identify and acquire suitable datasets for the task.
• Monitor processing pipelines to recognize bottlenecks.
• Process, organize, and store datasets for downstream use.

Objectives

• Analyze graph data with GPU-accelerated tools such as cuGraph.
• Optimize data science performance through GPU acceleration.
• Describe, follow, and execute CRISP-DM process steps.
• Use dependency management tools such as Docker and Conda to handle versioning conflicts.
• Determine optimal data type choices for feature columns.
• Design and implement benchmarks to compare framework performance.

Objectives

• Perform feature engineering for model development.
• Identify when data scale or workload profile requires acceleration.
• Run rapid experiments to balance accuracy and inference performance.
• Optimize machine learning hyperparameters.
• Train models and compare single-GPU versus multi-GPU strategies.
• Apply GPU memory optimization techniques such as batching and mixed precision.

Objectives

• Determine optimal data type choices for each feature.
• Assess and verify dataset memory footprint.
• Compare required memory against available device memory.
• Benchmark and optimize GPU-accelerated workflows.
• Deploy and monitor models in production.