Are Float, Int, Boolean correct types for V

[benikm91]

Currently, we use Float, Int, and Boolean for our Value Type V in Tensor.

val x: Tensor[T, Float] = Tensor(shape).fromArray(5f)

Using the existing Scala Types was somewhat arbitrary. And I see several issues:

• Tensor[T, Float] implies that it is float32, however, we track the precision in the internal dtype
• Tensor[T, Float] implies that Tensor[T, Double], Tensor[T, Byte] exists, but we want to track precision only internal.
• Scala has no unsigned type like UInt, which we should support (Natural Numbers).

I suggest moving to custom types: Floating, Integer, Bool

Lets use this issue as discussion for this.

[benikm91]

I like:

Real
Integer
Natural <-- new but already existing use case (image pixels 0..255)
Binary
(Complex) <-- we should add this later, when we have a use case.

[benikm91]

Just had an idea, we could do a type hierarchy for this. So one could be more specific if necessary.

So we have Int8, int16 that all extends Integer, etc.