Free NCA-ADS flashcards

How do you bring a pandas DataFrame onto the GPU so a cuDF merge with another cuDF DataFrame runs entirely on the device?

Convert the pandas DataFrame with cudf.from_pandas, which copies the host-side data into GPU memory and returns a cuDF DataFrame. With both frames on the GPU, the merge executes entirely on the device without further host-device transfers.

Activating the cudf.pandas accelerator automatically transfers pre-existing pandas DataFrames to the GPU without any conversion call.

The cudf.pandas accelerator intercepts pandas API calls on objects created after the module is activated. A pandas DataFrame that already exists in memory before activation is not retroactively moved to GPU without an explicit conversion step.

Why is stratified k-fold preferred over standard k-fold when a binary dataset has a small minority class (about 8%)?

With severe imbalance, random partitioning in standard k-fold may assign all minority samples to one fold, leaving some folds with no positive examples and making fold-level metrics meaningless. Stratified k-fold samples each class proportionally into every fold, preserving the overall class ratio and giving a consistent, stable evaluation signal across all splits.

Stratified k-fold helps with imbalance by not shuffling rows, preserving temporal sequence integrity that standard k-fold corrupts.

Shuffling and temporal ordering are concerns for time-series problems, not the class-imbalance problem. Stratified k-fold still shuffles rows, but does so within each class stratum, so it does not address a time-ordering concern. Its benefit for imbalance is preserving the class ratio in each fold.

In a dask_cudf workflow where a group-by aggregation does not run immediately, what triggers actual computation and dispatches work to the GPU workers?

dask_cudf is lazily evaluated: operations build a task graph but do not run until an action is called. Calling the compute method is the explicit trigger that instructs the Dask scheduler to execute all pending graph nodes across the available GPU workers and return a materialised cuDF result.

Printing a dask_cudf DataFrame forces Dask to resolve all partitions so the repr can display, reliably triggering full computation.

Printing a dask_cudf DataFrame displays a high-level metadata repr without executing the full task graph. Only a subset of metadata may be resolved, so it does not reliably trigger complete distributed computation across workers.

For salary data with a few extremely high earner outliers, which measure of central tendency best represents the typical employee?

The median. It splits the ordered distribution at the midpoint and is resistant to extreme values, so it is unaffected by how large the highest values are and reflects where most salaries actually fall. The mean would be dragged toward the tail.

The arithmetic mean is the best summary of typical salary because it incorporates every value in the dataset.

Using all data points is tempting, but outliers pull the mean far above most employees' salaries, making it unrepresentative of the typical worker. The median is preferred for skewed, outlier-affected data.