Accelerate Data Science on GPUs

1. What is the primary benefit of using cudf.pandas in your data science workflow?

   It provides new pandas functions not available in standard pandas

   It automatically accelerates pandas operations on GPU with zero code changes

   It reduces the memory footprint of pandas DataFrames

   It enables distributed computing across multiple machines

2. Which runtime environment on Google Cloud is recommended for teams that want fast setup, ease of use, and secure collaboration for data science workflows with Notebooks?

   Compute Engine

   Vertex AI Workbench

   Colab Enterprise

   Cloud Functions

3. In the context of GPU-accelerated machine learning, what does running the cuml.accel profilers help you identify?

   Memory leaks in your Python code

   Network latency issues

   CPU fallbacks and performance bottlenecks

   Syntax errors in your notebook

4. Which of the following is a best practice when using GPU acceleration for machine learning?

   Always use GPU for datasets of any size

   Use CSV format for fastest data loading

   Use Parquet format and prefer vectorized operations

   Process data row-by-row for maximum control

5. What parameter should you set in XGBoost to enable GPU acceleration?

   gpu=True

   accelerator='gpu'

   device='cuda'

   use_gpu=1

6. What is the key advantage of using cuML's estimators (e.g., cuml.ensemble.RandomForestClassifier) over their scikit-learn equivalents?

   They support more hyperparameter options than scikit-learn

   They train models on GPU, dramatically reducing fit and predict times on large datasets

   They automatically perform feature engineering before training

   They eliminate the need for hyperparameter tuning

7. When using GPU-accelerated XGBoost with cuDF DataFrames, what is a key benefit of passing cuDF data directly to xgb.train() or XGBClassifier.fit()?

   It automatically tunes the learning rate

   It avoids costly data transfers between GPU and CPU memory

   It enables training on multiple GPUs by default

   It converts categorical features to numeric automatically

8. What does the cuml.accel module allow you to do when working with scikit-learn code?

   Convert scikit-learn models to ONNX format automatically

   Run existing scikit-learn code on GPU with minimal or no code changes

   Distribute scikit-learn training across multiple CPU nodes

   Automatically select the best algorithm for your dataset

9. Both cuml.accel.profile and cuml.accel.line_profile are ways to profile your GPU-accelerated scikit-learn code. Which of the following are valid ways to invoke them?

   Add %%cuml.accel.profile to the top of a notebook cell.

   Add %%cuml.accel.line_profile to the top of a notebook cell.

   python -m cuml.accel --profile your_code.py

   python -m cuml.accel --line-profile your_code.py

   All of the above

Submit answers