Question

我正在尝试编写一些Haskell代码来表示现有库（HLearn）中的负二项分布。据我所知，HLearn要求模型参数（在这种情况下为负二项分布的r值）作为表示分布的类型类中的类型级自然数输入。我的问题是，当我尝试使用sing来访问参数的运行时值时，我得到一个类似的相关错误。

相关代码已注释

-- | Negative binomial distribution

{-# LANGUAGE
    GeneralizedNewtypeDeriving,
    TypeFamilies,
    DataKinds,
    KindSignatures #-}

module HLearn.Models.Distributions.Univariate.NegativeBinomial
    (NegativeBinomial
    )
    where

import GHC.TypeLits

import HLearn.Algebra
import HLearn.Models.Distributions.Common
import HLearn.Models.Distributions.Univariate.Internal.Moments

newtype NegativeBinomial (r::Nat) prob dp = NegativeBinomial { moments :: (Moments3 dp)     }
    deriving (Read, Show, Eq, Ord, Monoid, Group)

instance (Num dp) => HomTrainer (NegativeBinomial r prob dp) where
    type Datapoint (NegativeBinomial r prob dp) = dp
    train1dp dp = NegativeBinomial $ train1dp dp

instance (Num dp) => Probabilistic (NegativeBinomial r prob dp) where
    type Probability (NegativeBinomial r prob dp) = prob

-- The function in question
negbin_p :: (SingI r, Integral dp, Fractional prob) => NegativeBinomial r prob dp ->  prob
negbin_p (NegativeBinomial moments) = ((fromIntegral $ m0 moments)/(fromIntegral $ m1  moments)) * rval
    where
        rval = fromSing (sing :: Sing r) -- This is the code that is reported as causing errors

我得到的错误是：

src\HLearn\Models\Distributions\Univariate\NegativeBinomial.hs:33:24:
Could not deduce (SingE * (Kind *) prob)
  arising from a use of `fromSing'
from the context (SingI Nat r, Integral dp, Fractional prob)
  bound by the type signature for
             negbin_p :: (SingI Nat r, Integral dp, Fractional prob) =>
                         NegativeBinomial r prob dp -> prob
  at src\HLearn\Models\Distributions\Univariate\NegativeBinomial.hs:30:13-89
Possible fix:
  add an instance declaration for (SingE * (Kind *) prob)
In the expression: fromSing (sing :: Sing r)
In an equation for `rval': rval = fromSing (sing :: Sing r)
In an equation for `negbin_p':
    negbin_p (NegativeBinomial moments)
      = ((fromIntegral $ m0 moments) / (fromIntegral $ m1 moments))
        * rval
      where
          rval = fromSing (sing :: Sing r)

src\HLearn\Models\Distributions\Univariate\NegativeBinomial.hs:33:34:
Could not deduce (SingI * r1) arising from a use of `sing'
from the context (SingI Nat r, Integral dp, Fractional prob)
  bound by the type signature for
             negbin_p :: (SingI Nat r, Integral dp, Fractional prob) =>
                         NegativeBinomial r prob dp -> prob
  at src\HLearn\Models\Distributions\Univariate\NegativeBinomial.hs:30:13-89
Possible fix: add an instance declaration for (SingI * r1)
In the first argument of `fromSing', namely `(sing :: Sing r)'
In the expression: fromSing (sing :: Sing r)
In an equation for `rval': rval = fromSing (sing :: Sing r)

我做错了什么？

Answer 1

感谢luqui指出了ScopedTypeVariables。将其添加到语言编译指示后，更新的工作代码是相同的，除了：

negbin_p :: forall r dp prob . (SingI r, Integral dp, Fractional prob) => NegativeBinomial r prob dp -> prob
negbin_p (NegativeBinomial moments) = ((fromIntegral $ m0 moments)/(fromIntegral $ m1 moments)) * rval
    where
        rval = fromIntegral $ fromSing (sing :: Sing r)

Haskell从唱歌和唱歌打字/种类错误

1 个答案: