In lieu of trait arguments

Question

Here's what I want to write:

val alg10 = new GeneticAlgorithm[FlatDesign]
  with PhenotypeSize(10)
  with FlatDesignMutator
  with ChildrenReplaceParents(numChildren=2)
  with TargetFitnessFunction(targetPhenotype)
  with PopulationSize(40)
  with RunLength(numGenerations=100)

In other words, there are lots of constraints and parameters I'd like to set. For example, PhenotypeSize(10) has implications for the mutator and the fitness function. Abstract types ought to make it easy to implement/enforce all of those consistency constraints. Of course, traits can't take parameters, so this code won't work.

I love how composing traits lets you mix and match whatever functionality you need. The code above is so readable! With code like that, I could easily write a loop to try out the Cartesian product of lots and lots of variations to the algorithm.

I'm stuck, though, on finding a clean way to supply the parameters for those traits. Any suggestions?

som-snytt · Accepted Answer · 2014-05-12 07:21:01Z

2

Early definitions are described in Programming in Scala as "pre-initialized fields" or 5.1.6 of the spec and have the semantics you'd expect from trait parameters, namely, they are available in constructors.

val alg10 = new { val size = 10 } with GeneticAlgorithm[FlatDesign]
  with PhenotypeSize { val size: Int ; val psize = 2 * size }

This question must be a duplicate.

edited May 12, 2014 at 7:21

answered May 12, 2014 at 5:17

som-snytt

39.6k2 gold badges49 silver badges133 bronze badges

Sign up to request clarification or add additional context in comments.

3 Comments

Yuriy Over a year ago

In any case you can't mix the same trait with different constructor parameters. And it is a serious limitation of Cake Pattern and similar techniques.

Ben Kovitz Over a year ago

Thanks for the pointers to the specs! I didn't realize how readily available they were. I've been trial-and-erroring my way to the solution you posted (putting the early initializer first).

Ben Kovitz Over a year ago

Is there a way to make it easy to take the Cartesian product of many "traits"? "Ideal code" would look something like:

for (competition <- Seq(ChildrenReplaceParents, ParentsCompeteWithChildren; numChildren <- Seq(2, 3); fitnessFunc <- Seq(TargetFitnessFunc(target), SimpleFitnessFunc, RandomFitnessFunc); ...

Using traits is not a requirement; perhaps just passing a bunch of objects fits the problem better. Does Scala reflection provide a way to make this easy—or perhaps some feature I haven't even considered?

noziar · Accepted Answer · 2014-05-12 03:34:18Z

1

This may be what you are doing already, but just in case, if you define your traits like this:

trait PhenotypeSize {
    val size: Int
    ...
}
trait FlatDesignMutator { ... }
trait ChildrenReplaceParents {
    val numChildren: Int
    ...
}
...

Then your class will look like:

class GeneticAlgorithm[T]
    extends PhenotypeSize
    with FlatDesignMutator
    with ChildrenReplaceParents
    with ... {
    override val size: Int = 2
    override val numChildren: Int = 10
    ...
}

The compiler will check that you have given proper definitions for the members size, numChildren, and so on in your GeneticAlgorithm class implementation. This sounds reasonably nice and concise to me.

edited May 12, 2014 at 3:34

answered May 12, 2014 at 3:24

noziar

1,0578 silver badges11 bronze badges

1 Comment

Ben Kovitz Over a year ago

What you've written indeed looks a lot like what I've been hacking out this morning. :) It's not as pristinely readable as the ideal code above, but it's definitely workable. Please see my follow-up question to @som-snytt about walking through the Cartesian product of traits.

Collectives™ on Stack Overflow

In lieu of trait arguments

2 Answers 2

3 Comments

1 Comment

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

3 Comments

1 Comment

Your Answer

Sign up or log in

Post as a guest

Linked

Related