Codomain targeting.

This is a small experiment to see if targeting as a codomain leads to higher weight for a type. We need functions that act as distractions.

import $cp.bin.`provingground-core-jvm-9cc21d9ac6.fat.jar`
import provingground._ , interface._, HoTT._, learning._ 
repl.pprinter() = {
  val p = repl.pprinter()
  p.copy(
    additionalHandlers = p.additionalHandlers.orElse {
      translation.FansiShow.fansiHandler
    }
  )
}

import $cp.$                                              

import provingground._ , interface._, HoTT._, learning._ 

val M = "M" :: Type
val mul = "mul" :: (M ->: M ->: M)
val eqM = "eqM" ::(M ->: M ->: Type)
val m = "m" :: M
val n = "n" :: M

M: Typ[Term] = M
mul: Func[Term, Func[Term, Term]] = mul
eqM: Func[Term, Func[Term, Typ[Term]]] = eqM
m: Term = m
n: Term = n

val ts = TermState(FiniteDistribution.unif(mul, eqM, m , n), FiniteDistribution.unif(mul.typ, eqM.typ, M))

ts: TermState = TermState(
  FiniteDistribution(
    Vector(
      Weighted(mul, 0.25),
      Weighted(eqM, 0.25),
      Weighted(m, 0.25),
      Weighted(n, 0.25)
    )
  ),
  FiniteDistribution(
    Vector(
      Weighted((M → (M → M)), 0.3333333333333333),
      Weighted((M → (M → 𝒰 )), 0.3333333333333333),
      Weighted(M, 0.3333333333333333)
    )
  ),
  Vector(),
  FiniteDistribution(Vector()),
  FiniteDistribution(Vector()),
  Empty
)

val lp1 = LocalProver(ts, tg = TermGenParams()).noIsles
val lp2 = LocalProver(ts, tg = TermGenParams(typAsCodW = 0.1)).noIsles

lp1: LocalProver = LocalProver(
  TermState(
    FiniteDistribution(
      Vector(
        Weighted(mul, 0.25),
        Weighted(eqM, 0.25),
        Weighted(m, 0.25),
        Weighted(n, 0.25)
      )
    ),
    FiniteDistribution(
      Vector(
        Weighted((M → (M → M)), 0.3333333333333333),
        Weighted((M → (M → 𝒰 )), 0.3333333333333333),
        Weighted(M, 0.3333333333333333)
      )
    ),
    Vector(),
    FiniteDistribution(Vector()),
    FiniteDistribution(Vector()),
    Empty
  ),
  TermGenParams(
    0.1,
    0.1,
    0.1,
    0.0,
    0.0,
    0.05,
    0.05,
    0.0,
    0.0,
    0.0,
    0.0,
    0.0,
    0.3,
    0.7,
    0.5,
    0.0,
...
lp2: LocalProver = LocalProver(
  TermState(
    FiniteDistribution(
      Vector(
        Weighted(mul, 0.25),
        Weighted(eqM, 0.25),
        Weighted(m, 0.25),
        Weighted(n, 0.25)
      )
    ),
    FiniteDistribution(
      Vector(
        Weighted((M → (M → M)), 0.3333333333333333),
        Weighted((M → (M → 𝒰 )), 0.3333333333333333),
        Weighted(M, 0.3333333333333333)
      )
    ),
    Vector(),
    FiniteDistribution(Vector()),
    FiniteDistribution(Vector()),
    Empty
  ),
  TermGenParams(
    0.1,
    0.1,
    0.1,
    0.0,
    0.0,
    0.05,
    0.05,
    0.0,
    0.0,
    0.0,
    0.1,
    0.0,
    0.3,
    0.7,
    0.5,
    0.0,
...

val fd1T = lp1.varDist(TermGeneratorNodes.termsWithTyp(M))

fd1T: monix.eval.Task[FiniteDistribution[Term]] = Map(
  FlatMap(
    Async(<function2>, true, true, true),
    monix.eval.Task$$Lambda$2561/832550944@1882ecc8
  ),
  scala.Function1$$Lambda$335/412925308@16068149,
  1
)

val fd2T = lp2.varDist(TermGeneratorNodes.termsWithTyp(M))

fd2T: monix.eval.Task[FiniteDistribution[Term]] = Map(
  FlatMap(
    Async(<function2>, true, true, true),
    monix.eval.Task$$Lambda$2561/832550944@1a46478f
  ),
  scala.Function1$$Lambda$335/412925308@5eaa213b,
  1
)

import monix.execution.Scheduler.Implicits.global
fd1T.map(_.entropyVec).runToFuture

import monix.execution.Scheduler.Implicits.global

res6_1: monix.execution.CancelableFuture[Vector[Weighted[Term]]] = Success(
  Vector(
    Weighted(m, 2.343954401217361),
    Weighted(n, 2.343954401217361),
    Weighted(mul(n)(n), 3.0905340456590737),
    Weighted(mul(n)(m), 3.0905340456590737),
    Weighted(mul(m)(n), 3.0905340456590737),
    Weighted(mul(m)(m), 3.0905340456590737),
    Weighted(mul(n)(mul(m)(m)), 5.873311568780435),
    Weighted(mul(m)(mul(m)(m)), 5.873311568780435),
    Weighted(mul(m)(mul(n)(m)), 5.873311568780435),
    Weighted(mul(n)(mul(m)(n)), 5.873311568780435),
    Weighted(mul(n)(mul(n)(n)), 5.873311568780435),
    Weighted(mul(n)(mul(n)(m)), 5.873311568780435),
    Weighted(mul(m)(mul(n)(n)), 5.873311568780435),
    Weighted(mul(m)(mul(m)(n)), 5.873311568780435)
  )
)

fd2T.map(_.entropyVec).runToFuture

res7: monix.execution.CancelableFuture[Vector[Weighted[Term]]] = Success(
  Vector(
    Weighted(mul(n)(n), 3.00286602433109),
    Weighted(mul(n)(m), 3.00286602433109),
    Weighted(mul(m)(n), 3.00286602433109),
    Weighted(mul(m)(m), 3.00286602433109),
    Weighted(m, 3.0274807364221066),
    Weighted(n, 3.0274807364221066),
    Weighted(mul(n)(mul(m)(m)), 4.967423556425984),
    Weighted(mul(m)(mul(m)(m)), 4.967423556425984),
    Weighted(mul(m)(mul(n)(m)), 4.967423556425984),
    Weighted(mul(n)(mul(m)(n)), 4.967423556425984),
    Weighted(mul(n)(mul(n)(n)), 4.967423556425984),
    Weighted(mul(n)(mul(n)(m)), 4.967423556425984),
    Weighted(mul(m)(mul(n)(n)), 4.967423556425984),
    Weighted(mul(m)(mul(m)(n)), 4.967423556425984)
  )
)

Unify.targetCodomain(mul, M)

res8: Option[Term] = Some(($dxlc : M) ↦ ($dxlb : M) ↦ mul($dxlb)($dxlc))

Conclusion

We see that with strong targeting we get a much sharper distribution. We should do this for better instantiation in the Achal problem.