Big Data Analytics MCQs

A clear signature helps users and tools understand an API.
It also keeps internal implementation changes from altering the exposed type unexpectedly.

The scheduler launches tasks on executors.
Parallelism within a stage is largely determined by partition count.

The zero value must behave as an identity for correct parallel folding.
It may be applied more than once in a distributed computation.

Checkpointing helps when lineage becomes long or cyclic algorithms run many iterations.
It requires materialization to durable storage.

Structured formats can avoid reading unused columns.
This lowers I/O, memory use, and CPU work.

It evaluates score-based binary classification metrics such as area under ROC or PR.
The metric should match the application's priorities.

Spark supports batch, SQL, machine learning, and streaming workloads.
It can run on clusters managed by YARN, Kubernetes, or standalone mode.

Operations such as map and filter are usually narrow.
They can often be pipelined without shuffling data.

aggregate uses one function within partitions and another across partition results.
Its zero value can have a different type from RDD elements.

Cached blocks may be lost and rebuilt from lineage.
Checkpoint data becomes a new reliable starting point.

Columnar storage reads only relevant fields and compresses similar values well.
Metadata can also help skip unnecessary data.

Silhouette compares within-cluster similarity with separation from other clusters.
Higher values generally indicate clearer assignments.