Instruction Implementation Meta Issue

[fschueler]

This issue tracks the status of instructions. Please insert the instruction that you're working on and your name into the table and don't forget to mark it as finished once you're done.

Instruction	Assignee	Status	Comment
TSMMFLInstruction	@fschueler	✅
ReblockFLInstruction	@fschueler	✅
CSVReblockFLInstruction	@fschueler	✅
MatrixScalarFLInstruction	@carabolic	✅	Should work for all supported scalar operations; test wit MLContext is missing
WriteFLInstruction	@fschueler	✅
RandFLInstruction	@fschueler	✅
MapmmFLInstruction	@carabolic	✅
MapmmChainFLInstruction	@carabolic	✅
ArithmeticBinaryFLInstruction	@fschueler	✅
MatrixMatrixArithmeticFLInstruction	@fschueler	✅
AggregateUnaryFLInstruction	@fschueler	✅

Symbol	Explanation
✅	completed
🚧	in progress
❌	will/can not be implemented

[carabolic]

I see a lot of times Spark instructions are manually optimized to use a broadcast/ map-side join. I think this optimization is not needed for Flink. I think we should always use the Flink operator with the best semantics and give the Flink optimizer the enough meta information to find the perfect plan. For instance the matrix multiplication (Mapmm and MapmmChain) could be extended to use Flink's JoinHints.