When to use RDDs, Datasets, and DataFrames?

The following table describes the scenarios in which RDDs, Datasets, or DataFrames are to be used:

Scenario	What to use?
Use of the Python programming language	RDDs or DataFrames
Use of the R programming language	DataFrames
Use of the Java or Scala programming languages	RDDs, Datasets, or DataFrames
Unstructured data such as images and videos	RDDs
Use of low level transformations, actions, and controls data flow programmatically	RDDs
Use of high-level domain-specific APIs	Datasets and DataFrames
Use of functional programming constructs to process data	RDDs
Use of higher level expressions including SQLs	Datasets and DataFrames
Imposing structure is not needed and low-level optimizations are not needed	RDDs
High compile time safety and rich optimizations	Datasets
No compile time safety and rich optimizations are needed	DataFrames
Unification is needed across Spark libraries	Datasets or DataFrames