Merge pull request #232 from quantified-uncertainty/testing-discipline-algebraic-operations

Testing discipline: algebraic operations
This commit is contained in:
Ozzie Gooen 2022-04-13 14:00:58 -04:00 committed by GitHub
commit bd10a0bbf8
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
16 changed files with 629 additions and 50 deletions

View File

@ -18,11 +18,9 @@ let {
triangularDist,
exponentialDist,
} = module(GenericDist_Fixtures)
let mkNormal = (mean, stdev) => GenericDist_Types.Symbolic(#Normal({mean: mean, stdev: stdev}))
let {toFloat, toDist, toString, toError} = module(DistributionOperation.Output)
let {toFloat, toDist, toString, toError, fmap} = module(DistributionOperation.Output)
let {run} = module(DistributionOperation)
let {fmap} = module(DistributionOperation.Output)
let run = run(~env)
let outputMap = fmap(~env)
let toExt: option<'a> => 'a = E.O.toExt(

View File

@ -11,3 +11,4 @@ let triangularDist: GenericDist_Types.genericDist = Symbolic(
)
let exponentialDist: GenericDist_Types.genericDist = Symbolic(#Exponential({rate: 2.0}))
let uniformDist: GenericDist_Types.genericDist = Symbolic(#Uniform({low: 9.0, high: 10.0}))
let floatDist: GenericDist_Types.genericDist = Symbolic(#Float(1e1))

View File

@ -0,0 +1,368 @@
/*
This file is like a half measure between one-off unit tests and proper invariant validation.
As such, I'm not that excited about it, though it does provide some structure and will alarm us
when things substantially change.
Also, there are some open comments in https://github.com/quantified-uncertainty/squiggle/pull/232 that haven't been addressed.
*/
open Jest
open Expect
open TestHelpers
let {
normalDist5, // mean=5, stdev=2
normalDist10, // mean=10, stdev=2
normalDist20, // mean=20, stdev=2
normalDist, // mean=5; stdev=2
uniformDist, // low=9; high=10
betaDist, // alpha=2; beta=5
lognormalDist, // mu=0; sigma=1
cauchyDist, // local=1; scale=1
triangularDist, // low=1; medium=2; high=3;
exponentialDist, // rate=2
} = module(GenericDist_Fixtures)
let {
algebraicAdd,
algebraicMultiply,
algebraicDivide,
algebraicSubtract,
algebraicLogarithm,
algebraicPower,
} = module(DistributionOperation.Constructors)
let algebraicAdd = algebraicAdd(~env)
let algebraicMultiply = algebraicMultiply(~env)
let algebraicDivide = algebraicDivide(~env)
let algebraicSubtract = algebraicSubtract(~env)
let algebraicLogarithm = algebraicLogarithm(~env)
let algebraicPower = algebraicPower(~env)
describe("(Algebraic) addition of distributions", () => {
describe("mean", () => {
test("normal(mean=5) + normal(mean=20)", () => {
normalDist5
->algebraicAdd(normalDist20)
->E.R2.fmap(GenericDist_Types.Constructors.UsingDists.mean)
->E.R2.fmap(run)
->E.R2.fmap(toFloat)
->E.R.toExn
->expect
->toBe(Some(2.5e1))
})
test("uniform(low=9, high=10) + beta(alpha=2, beta=5)", () => {
// let uniformMean = (9.0 +. 10.0) /. 2.0
// let betaMean = 1.0 /. (1.0 +. 5.0 /. 2.0)
let received =
uniformDist
->algebraicAdd(betaDist)
->E.R2.fmap(GenericDist_Types.Constructors.UsingDists.mean)
->E.R2.fmap(run)
->E.R2.fmap(toFloat)
->E.R.toExn
switch received {
| None => "algebraicAdd has"->expect->toBe("failed")
// This is nondeterministic, we could be in a situation where ci fails but you click rerun and it passes, which is bad.
// sometimes it works with ~digits=2.
| Some(x) => x->expect->toBeSoCloseTo(0.01927225696028752, ~digits=1) // (uniformMean +. betaMean)
}
})
test("beta(alpha=2, beta=5) + uniform(low=9, high=10)", () => {
// let uniformMean = (9.0 +. 10.0) /. 2.0
// let betaMean = 1.0 /. (1.0 +. 5.0 /. 2.0)
let received =
betaDist
->algebraicAdd(uniformDist)
->E.R2.fmap(GenericDist_Types.Constructors.UsingDists.mean)
->E.R2.fmap(run)
->E.R2.fmap(toFloat)
->E.R.toExn
switch received {
| None => "algebraicAdd has"->expect->toBe("failed")
// This is nondeterministic, we could be in a situation where ci fails but you click rerun and it passes, which is bad.
// sometimes it works with ~digits=2.
| Some(x) => x->expect->toBeSoCloseTo(0.019275414920485248, ~digits=1) // (uniformMean +. betaMean)
}
})
})
describe("pdf", () => {
// TEST IS WRONG. SEE STDEV ADDITION EXPRESSION.
testAll(
"(normal(mean=5) + normal(mean=5)).pdf (imprecise)",
list{8e0, 1e1, 1.2e1, 1.4e1},
x => {
let received =
normalDist10 // this should be normal(10, sqrt(8))
->Ok
->E.R2.fmap(d => GenericDist_Types.Constructors.UsingDists.pdf(d, x))
->E.R2.fmap(run)
->E.R2.fmap(toFloat)
->E.R.toOption
->E.O.flatten
let calculated =
normalDist5
->algebraicAdd(normalDist5)
->E.R2.fmap(d => GenericDist_Types.Constructors.UsingDists.pdf(d, x))
->E.R2.fmap(run)
->E.R2.fmap(toFloat)
->E.R.toOption
->E.O.flatten
switch received {
| None =>
"this branch occurs when the dispatch to Jstat on trusted input fails."
->expect
->toBe("never")
| Some(x) =>
switch calculated {
| None => "algebraicAdd has"->expect->toBe("failed")
| Some(y) => x->expect->toBeSoCloseTo(y, ~digits=0)
}
}
},
)
test("(normal(mean=10) + normal(mean=10)).pdf(1.9e1)", () => {
let received =
normalDist20
->Ok
->E.R2.fmap(d => GenericDist_Types.Constructors.UsingDists.pdf(d, 1.9e1))
->E.R2.fmap(run)
->E.R2.fmap(toFloat)
->E.R.toOption
->E.O.flatten
let calculated =
normalDist10
->algebraicAdd(normalDist10)
->E.R2.fmap(d => GenericDist_Types.Constructors.UsingDists.pdf(d, 1.9e1))
->E.R2.fmap(run)
->E.R2.fmap(toFloat)
->E.R.toOption
->E.O.flatten
switch received {
| None =>
"this branch occurs when the dispatch to Jstat on trusted input fails."
->expect
->toBe("never")
| Some(x) =>
switch calculated {
| None => "algebraicAdd has"->expect->toBe("failed")
| Some(y) => x->expect->toBeSoCloseTo(y, ~digits=1)
}
}
})
test("(uniform(low=9, high=10) + beta(alpha=2, beta=5)).pdf(10)", () => {
let received =
uniformDist
->algebraicAdd(betaDist)
->E.R2.fmap(d => GenericDist_Types.Constructors.UsingDists.pdf(d, 1e1))
->E.R2.fmap(run)
->E.R2.fmap(toFloat)
->E.R.toExn
switch received {
| None => "algebraicAdd has"->expect->toBe("failed")
// This is nondeterministic, we could be in a situation where ci fails but you click rerun and it passes, which is bad.
// sometimes it works with ~digits=4.
| Some(x) => x->expect->toBeSoCloseTo(0.001978994877226945, ~digits=3)
}
})
test("(beta(alpha=2, beta=5) + uniform(low=9, high=10)).pdf(10)", () => {
let received =
betaDist
->algebraicAdd(uniformDist)
->E.R2.fmap(d => GenericDist_Types.Constructors.UsingDists.pdf(d, 1e1))
->E.R2.fmap(run)
->E.R2.fmap(toFloat)
->E.R.toExn
switch received {
| None => "algebraicAdd has"->expect->toBe("failed")
// This is nondeterministic, we could be in a situation where ci fails but you click rerun and it passes, which is bad.
// sometimes it works with ~digits=4.
| Some(x) => x->expect->toBeSoCloseTo(0.001978994877226945, ~digits=3)
}
})
})
describe("cdf", () => {
testAll("(normal(mean=5) + normal(mean=5)).cdf (imprecise)", list{6e0, 8e0, 1e1, 1.2e1}, x => {
let received =
normalDist10
->Ok
->E.R2.fmap(d => GenericDist_Types.Constructors.UsingDists.cdf(d, x))
->E.R2.fmap(run)
->E.R2.fmap(toFloat)
->E.R.toOption
->E.O.flatten
let calculated =
normalDist5
->algebraicAdd(normalDist5)
->E.R2.fmap(d => GenericDist_Types.Constructors.UsingDists.cdf(d, x))
->E.R2.fmap(run)
->E.R2.fmap(toFloat)
->E.R.toOption
->E.O.flatten
switch received {
| None =>
"this branch occurs when the dispatch to Jstat on trusted input fails."
->expect
->toBe("never")
| Some(x) =>
switch calculated {
| None => "algebraicAdd has"->expect->toBe("failed")
| Some(y) => x->expect->toBeSoCloseTo(y, ~digits=0)
}
}
})
test("(normal(mean=10) + normal(mean=10)).cdf(1.25e1)", () => {
let received =
normalDist20
->Ok
->E.R2.fmap(d => GenericDist_Types.Constructors.UsingDists.cdf(d, 1.25e1))
->E.R2.fmap(run)
->E.R2.fmap(toFloat)
->E.R.toOption
->E.O.flatten
let calculated =
normalDist10
->algebraicAdd(normalDist10)
->E.R2.fmap(d => GenericDist_Types.Constructors.UsingDists.cdf(d, 1.25e1))
->E.R2.fmap(run)
->E.R2.fmap(toFloat)
->E.R.toOption
->E.O.flatten
switch received {
| None =>
"this branch occurs when the dispatch to Jstat on trusted input fails."
->expect
->toBe("never")
| Some(x) =>
switch calculated {
| None => "algebraicAdd has"->expect->toBe("failed")
| Some(y) => x->expect->toBeSoCloseTo(y, ~digits=2)
}
}
})
test("(uniform(low=9, high=10) + beta(alpha=2, beta=5)).cdf(10)", () => {
let received =
uniformDist
->algebraicAdd(betaDist)
->E.R2.fmap(d => GenericDist_Types.Constructors.UsingDists.cdf(d, 1e1))
->E.R2.fmap(run)
->E.R2.fmap(toFloat)
->E.R.toExn
switch received {
| None => "algebraicAdd has"->expect->toBe("failed")
// This is nondeterministic, we could be in a situation where ci fails but you click rerun and it passes, which is bad.
// sometimes it works with ~digits=4.
| Some(x) => x->expect->toBeSoCloseTo(0.0013961779932477507, ~digits=3)
}
})
test("(beta(alpha=2, beta=5) + uniform(low=9, high=10)).cdf(10)", () => {
let received =
betaDist
->algebraicAdd(uniformDist)
->E.R2.fmap(d => GenericDist_Types.Constructors.UsingDists.cdf(d, 1e1))
->E.R2.fmap(run)
->E.R2.fmap(toFloat)
->E.R.toExn
switch received {
| None => "algebraicAdd has"->expect->toBe("failed")
// This is nondeterministic, we could be in a situation where ci fails but you click rerun and it passes, which is bad.
// sometimes it works with ~digits=4.
| Some(x) => x->expect->toBeSoCloseTo(0.001388898111625753, ~digits=3)
}
})
})
describe("inv", () => {
testAll("(normal(mean=5) + normal(mean=5)).inv (imprecise)", list{5e-2, 4.2e-3, 9e-3}, x => {
let received =
normalDist10
->Ok
->E.R2.fmap(d => GenericDist_Types.Constructors.UsingDists.inv(d, x))
->E.R2.fmap(run)
->E.R2.fmap(toFloat)
->E.R.toOption
->E.O.flatten
let calculated =
normalDist5
->algebraicAdd(normalDist5)
->E.R2.fmap(d => GenericDist_Types.Constructors.UsingDists.inv(d, x))
->E.R2.fmap(run)
->E.R2.fmap(toFloat)
->E.R.toOption
->E.O.flatten
switch received {
| None =>
"this branch occurs when the dispatch to Jstat on trusted input fails."
->expect
->toBe("never")
| Some(x) =>
switch calculated {
| None => "algebraicAdd has"->expect->toBe("failed")
| Some(y) => x->expect->toBeSoCloseTo(y, ~digits=-1)
}
}
})
test("(normal(mean=10) + normal(mean=10)).inv(1e-1)", () => {
let received =
normalDist20
->Ok
->E.R2.fmap(d => GenericDist_Types.Constructors.UsingDists.inv(d, 1e-1))
->E.R2.fmap(run)
->E.R2.fmap(toFloat)
->E.R.toOption
->E.O.flatten
let calculated =
normalDist10
->algebraicAdd(normalDist10)
->E.R2.fmap(d => GenericDist_Types.Constructors.UsingDists.inv(d, 1e-1))
->E.R2.fmap(run)
->E.R2.fmap(toFloat)
->E.R.toOption
->E.O.flatten
switch received {
| None =>
"this branch occurs when the dispatch to Jstat on trusted input fails."
->expect
->toBe("never")
| Some(x) =>
switch calculated {
| None => "algebraicAdd has"->expect->toBe("failed")
| Some(y) => x->expect->toBeSoCloseTo(y, ~digits=-1)
}
}
})
test("(uniform(low=9, high=10) + beta(alpha=2, beta=5)).inv(2e-2)", () => {
let received =
uniformDist
->algebraicAdd(betaDist)
->E.R2.fmap(d => GenericDist_Types.Constructors.UsingDists.inv(d, 2e-2))
->E.R2.fmap(run)
->E.R2.fmap(toFloat)
->E.R.toExn
switch received {
| None => "algebraicAdd has"->expect->toBe("failed")
// This is nondeterministic, we could be in a situation where ci fails but you click rerun and it passes, which is bad.
// sometimes it works with ~digits=2.
| Some(x) => x->expect->toBeSoCloseTo(10.927078217530806, ~digits=0)
}
})
test("(beta(alpha=2, beta=5) + uniform(low=9, high=10)).inv(2e-2)", () => {
let received =
betaDist
->algebraicAdd(uniformDist)
->E.R2.fmap(d => GenericDist_Types.Constructors.UsingDists.inv(d, 2e-2))
->E.R2.fmap(run)
->E.R2.fmap(toFloat)
->E.R.toExn
switch received {
| None => "algebraicAdd has"->expect->toBe("failed")
// This is nondeterministic, we could be in a situation where ci fails but you click rerun and it passes, which is bad.
// sometimes it works with ~digits=2.
| Some(x) => x->expect->toBeSoCloseTo(10.915396627014363, ~digits=0)
}
})
})
})

View File

@ -0,0 +1,140 @@
/*
This is the most basic file in our invariants family of tests.
See document in https://github.com/quantified-uncertainty/squiggle/pull/238 for details
Note: digits parameter should be higher than -4.
*/
open Jest
open Expect
open TestHelpers
let {
algebraicAdd,
algebraicMultiply,
algebraicDivide,
algebraicSubtract,
algebraicLogarithm,
algebraicPower,
} = module(DistributionOperation.Constructors)
let algebraicAdd = algebraicAdd(~env)
let algebraicMultiply = algebraicMultiply(~env)
let algebraicDivide = algebraicDivide(~env)
let algebraicSubtract = algebraicSubtract(~env)
let algebraicLogarithm = algebraicLogarithm(~env)
let algebraicPower = algebraicPower(~env)
describe("Mean", () => {
let digits = -4
let mean = GenericDist_Types.Constructors.UsingDists.mean
let runMean: result<DistributionTypes.genericDist, DistributionTypes.error> => float = distR => {
distR
->E.R2.fmap(mean)
->E.R2.fmap(run)
->E.R2.fmap(toFloat)
->E.R.toExn
->E.O2.toExn("Shouldn't see this because we trust testcase input")
}
let impossiblePath: string => assertion = algebraicOp =>
`${algebraicOp} has`->expect->toEqual("failed")
let distributions = list{
normalMake(0.0, 1e0),
betaMake(2e0, 4e0),
exponentialMake(1.234e0),
uniformMake(7e0, 1e1),
// cauchyMake(1e0, 1e0),
lognormalMake(1e0, 1e0),
triangularMake(1e0, 1e1, 5e1),
Ok(floatMake(1e1)),
}
let combinations = E.L.combinations2(distributions)
let zipDistsDists = E.L.zip(distributions, distributions)
let testOperationMean = (
distOp: (DistributionTypes.genericDist, DistributionTypes.genericDist) => result<DistributionTypes.genericDist, DistributionTypes.error>,
description: string,
floatOp: (float, float) => float,
dist1': result<SymbolicDistTypes.symbolicDist, string>,
dist2': result<SymbolicDistTypes.symbolicDist, string>
) => {
let dist1 = dist1'->E.R2.fmap(x=>DistributionTypes.Symbolic(x))->E.R2.fmap2(s=>DistributionTypes.Other(s))
let dist2 = dist2'->E.R2.fmap(x=>DistributionTypes.Symbolic(x))->E.R2.fmap2(s=>DistributionTypes.Other(s))
let received =
E.R.liftJoin2(distOp, dist1, dist2)
->E.R2.fmap(mean)
->E.R2.fmap(run)
->E.R2.fmap(toFloat)
let expected = floatOp(runMean(dist1), runMean(dist2))
switch received {
| Error(err) => impossiblePath(description)
| Ok(x) =>
switch x {
| None => impossiblePath(description)
| Some(x) => x->expect->toBeSoCloseTo(expected, ~digits)
}
}
}
describe("addition", () => {
let testAdditionMean = testOperationMean(algebraicAdd, "algebraicAdd", \"+.")
testAll("homogeneous addition", zipDistsDists, dists => {
let (dist1, dist2) = dists
testAdditionMean(dist1, dist2)
})
testAll("heterogeneous addition (1)", combinations, dists => {
let (dist1, dist2) = dists
testAdditionMean(dist1, dist2)
})
testAll("heterogeneous addition (commuted of 1 (or; 2))", combinations, dists => {
let (dist1, dist2) = dists
testAdditionMean(dist2, dist1)
})
})
describe("subtraction", () => {
let testSubtractionMean = testOperationMean(algebraicSubtract, "algebraicSubtract", \"-.")
testAll("homogeneous subtraction", zipDistsDists, dists => {
let (dist1, dist2) = dists
testSubtractionMean(dist1, dist2)
})
testAll("heterogeneous subtraction (1)", combinations, dists => {
let (dist1, dist2) = dists
testSubtractionMean(dist1, dist2)
})
testAll("heterogeneous subtraction (commuted of 1 (or; 2))", combinations, dists => {
let (dist1, dist2) = dists
testSubtractionMean(dist2, dist1)
})
})
describe("multiplication", () => {
let testMultiplicationMean = testOperationMean(algebraicMultiply, "algebraicMultiply", \"*.")
testAll("homogeneous subtraction", zipDistsDists, dists => {
let (dist1, dist2) = dists
testMultiplicationMean(dist1, dist2)
})
testAll("heterogeneoous subtraction (1)", combinations, dists => {
let (dist1, dist2) = dists
testMultiplicationMean(dist1, dist2)
})
testAll("heterogeneoous subtraction (commuted of 1 (or; 2))", combinations, dists => {
let (dist1, dist2) = dists
testMultiplicationMean(dist2, dist1)
})
})
})

View File

@ -2,14 +2,6 @@ open Jest
open Expect
open TestHelpers
// TODO: use Normal.make (etc.), but preferably after the new validation dispatch is in.
let mkNormal = (mean, stdev) => GenericDist_Types.Symbolic(#Normal({mean: mean, stdev: stdev}))
let mkBeta = (alpha, beta) => GenericDist_Types.Symbolic(#Beta({alpha: alpha, beta: beta}))
let mkExponential = rate => GenericDist_Types.Symbolic(#Exponential({rate: rate}))
let mkUniform = (low, high) => GenericDist_Types.Symbolic(#Uniform({low: low, high: high}))
let mkCauchy = (local, scale) => GenericDist_Types.Symbolic(#Cauchy({local: local, scale: scale}))
let mkLognormal = (mu, sigma) => GenericDist_Types.Symbolic(#Lognormal({mu: mu, sigma: sigma}))
describe("mixture", () => {
testAll(
"fair mean of two normal distributions",

View File

@ -3,7 +3,7 @@ open Expect
open TestHelpers
// TODO: use Normal.make (but preferably after teh new validation dispatch is in)
let mkNormal = (mean, stdev) => GenericDist_Types.Symbolic(#Normal({mean: mean, stdev: stdev}))
let mkNormal = (mean, stdev) => DistributionTypes.Symbolic(#Normal({mean: mean, stdev: stdev}))
describe("(Symbolic) normalize", () => {
testAll("has no impact on normal distributions", list{-1e8, -1e-2, 0.0, 1e-4, 1e16}, mean => {
@ -28,16 +28,16 @@ describe("(Symbolic) mean", () => {
testAll("of exponential distributions", list{1e-7, 2.0, 10.0, 100.0}, rate => {
let meanValue = run(
FromDist(ToFloat(#Mean), GenericDist_Types.Symbolic(#Exponential({rate: rate}))),
FromDist(ToFloat(#Mean), DistributionTypes.Symbolic(#Exponential({rate: rate}))),
)
meanValue->unpackFloat->expect->toBeCloseTo(1.0 /. rate) // https://en.wikipedia.org/wiki/Exponential_distribution#Mean,_variance,_moments,_and_median
})
test("of a cauchy distribution", () => {
let meanValue = run(
FromDist(ToFloat(#Mean), GenericDist_Types.Symbolic(#Cauchy({local: 1.0, scale: 1.0}))),
FromDist(ToFloat(#Mean), DistributionTypes.Symbolic(#Cauchy({local: 1.0, scale: 1.0}))),
)
meanValue->unpackFloat->expect->toBeCloseTo(2.01868297874546)
meanValue->unpackFloat->expect->toBeSoCloseTo(1.0098094001641797, ~digits=5)
//-> toBe(GenDistError(Other("Cauchy distributions may have no mean value.")))
})
@ -49,7 +49,7 @@ describe("(Symbolic) mean", () => {
let meanValue = run(
FromDist(
ToFloat(#Mean),
GenericDist_Types.Symbolic(#Triangular({low: low, medium: medium, high: high})),
DistributionTypes.Symbolic(#Triangular({low: low, medium: medium, high: high})),
),
)
meanValue->unpackFloat->expect->toBeCloseTo((low +. medium +. high) /. 3.0) // https://www.statology.org/triangular-distribution/
@ -63,7 +63,7 @@ describe("(Symbolic) mean", () => {
tup => {
let (alpha, beta) = tup
let meanValue = run(
FromDist(ToFloat(#Mean), GenericDist_Types.Symbolic(#Beta({alpha: alpha, beta: beta}))),
FromDist(ToFloat(#Mean), DistributionTypes.Symbolic(#Beta({alpha: alpha, beta: beta}))),
)
meanValue->unpackFloat->expect->toBeCloseTo(1.0 /. (1.0 +. beta /. alpha)) // https://en.wikipedia.org/wiki/Beta_distribution#Mean
},
@ -72,7 +72,7 @@ describe("(Symbolic) mean", () => {
// TODO: When we have our theory of validators we won't want this to be NaN but to be an error.
test("of beta(0, 0)", () => {
let meanValue = run(
FromDist(ToFloat(#Mean), GenericDist_Types.Symbolic(#Beta({alpha: 0.0, beta: 0.0}))),
FromDist(ToFloat(#Mean), DistributionTypes.Symbolic(#Beta({alpha: 0.0, beta: 0.0}))),
)
meanValue->unpackFloat->expect->ExpectJs.toBeFalsy
})
@ -83,7 +83,7 @@ describe("(Symbolic) mean", () => {
tup => {
let (mu, sigma) = tup
let meanValue = run(
FromDist(ToFloat(#Mean), GenericDist_Types.Symbolic(#Lognormal({mu: mu, sigma: sigma}))),
FromDist(ToFloat(#Mean), DistributionTypes.Symbolic(#Lognormal({mu: mu, sigma: sigma}))),
)
meanValue->unpackFloat->expect->toBeCloseTo(Js.Math.exp(mu +. sigma ** 2.0 /. 2.0)) // https://brilliant.org/wiki/log-normal-distribution/
},
@ -95,14 +95,14 @@ describe("(Symbolic) mean", () => {
tup => {
let (low, high) = tup
let meanValue = run(
FromDist(ToFloat(#Mean), GenericDist_Types.Symbolic(#Uniform({low: low, high: high}))),
FromDist(ToFloat(#Mean), DistributionTypes.Symbolic(#Uniform({low: low, high: high}))),
)
meanValue->unpackFloat->expect->toBeCloseTo((low +. high) /. 2.0) // https://en.wikipedia.org/wiki/Continuous_uniform_distribution#Moments
},
)
test("of a float", () => {
let meanValue = run(FromDist(ToFloat(#Mean), GenericDist_Types.Symbolic(#Float(7.7))))
let meanValue = run(FromDist(ToFloat(#Mean), DistributionTypes.Symbolic(#Float(7.7))))
meanValue->unpackFloat->expect->toBeCloseTo(7.7)
})
})

View File

@ -11,17 +11,33 @@ let {toFloat, toDist, toString, toError, fmap} = module(DistributionOperation.Ou
let fnImage = (theFn, inps) => Js.Array.map(theFn, inps)
let env: DistributionOperation.env = {
sampleCount: 100,
xyPointLength: 100,
sampleCount: 10000,
xyPointLength: 1000,
}
let run = DistributionOperation.run(~env)
let outputMap = fmap(~env)
let unreachableInTestFileMessage = "Should be impossible to reach (This error is in test file)"
let toExtFloat: option<float> => float = E.O.toExt(unreachableInTestFileMessage)
let toExtDist: option<GenericDist_Types.genericDist> => GenericDist_Types.genericDist = E.O.toExt(
let toExtDist: option<DistributionTypes.genericDist> => DistributionTypes.genericDist = E.O.toExt(
unreachableInTestFileMessage,
)
// let toExt: option<'a> => 'a = E.O.toExt(unreachableInTestFileMessage)
let unpackFloat = x => x->toFloat->toExtFloat
let unpackDist = y => y->toDist->toExtDist
let mkNormal = (mean, stdev) => DistributionTypes.Symbolic(#Normal({mean: mean, stdev: stdev}))
let mkBeta = (alpha, beta) => DistributionTypes.Symbolic(#Beta({alpha: alpha, beta: beta}))
let mkExponential = rate => DistributionTypes.Symbolic(#Exponential({rate: rate}))
let mkUniform = (low, high) => DistributionTypes.Symbolic(#Uniform({low: low, high: high}))
let mkCauchy = (local, scale) => DistributionTypes.Symbolic(#Cauchy({local: local, scale: scale}))
let mkLognormal = (mu, sigma) => DistributionTypes.Symbolic(#Lognormal({mu: mu, sigma: sigma}))
let normalMake = SymbolicDist.Normal.make
let betaMake = SymbolicDist.Beta.make
let exponentialMake = SymbolicDist.Exponential.make
let uniformMake = SymbolicDist.Uniform.make
let cauchyMake = SymbolicDist.Cauchy.make
let lognormalMake = SymbolicDist.Lognormal.make
let triangularMake = SymbolicDist.Triangular.make
let floatMake = SymbolicDist.Float.make

View File

@ -0,0 +1,10 @@
open Jest
open Expect
describe("E.L.combinations2", () => {
test("size three", () => {
E.L.combinations2(list{"alice", "bob", "eve"})
->expect
->toEqual(list{("alice", "bob"), ("alice", "eve"), ("bob", "eve")})
})
})

View File

@ -10,6 +10,7 @@
"test:reducer": "jest --testPathPattern '.*__tests__/Reducer.*'",
"test": "jest",
"test:watch": "jest --watchAll",
"test:quick": "jest --modulePathIgnorePatterns=__tests__/Distributions/Invariants/*",
"coverage": "rm -f *.coverage; yarn clean; BISECT_ENABLE=yes yarn build; yarn test; bisect-ppx-report html",
"coverage:ci": "yarn clean; BISECT_ENABLE=yes yarn build; yarn test; bisect-ppx-report send-to Codecov",
"lint:rescript": "./lint.sh",

View File

@ -1,6 +1,6 @@
type functionCallInfo = GenericDist_Types.Operation.genericFunctionCallInfo
type genericDist = GenericDist_Types.genericDist
type error = GenericDist_Types.error
type genericDist = DistributionTypes.genericDist
type error = DistributionTypes.error
// TODO: It could be great to use a cache for some calculations (basically, do memoization). Also, better analytics/tracking could go a long way.

View File

@ -1,12 +1,15 @@
@genType
type genericDist =
| PointSet(PointSetTypes.pointSetDist)
| SampleSet(array<float>)
| SampleSet(SampleSetDist.t)
| Symbolic(SymbolicDistTypes.symbolicDist)
@genType
type error =
| NotYetImplemented
| Unreachable
| DistributionVerticalShiftIsInvalid
| ArgumentError(string)
| Other(string)
module Operation = {

View File

@ -1,6 +1,6 @@
//TODO: multimodal, add interface, test somehow, track performance, refactor sampleSet, refactor ASTEvaluator.res.
type t = GenericDist_Types.genericDist
type error = GenericDist_Types.error
type t = DistributionTypes.genericDist
type error = DistributionTypes.error
type toPointSetFn = t => result<PointSetTypes.pointSetDist, error>
type toSampleSetFn = t => result<SampleSetDist.t, error>
type scaleMultiplyFn = (t, float) => result<t, error>
@ -115,7 +115,7 @@ module Truncate = {
| Some(r) => Ok(r)
| None =>
toPointSetFn(t)->E.R2.fmap(t => {
GenericDist_Types.PointSet(PointSetDist.T.truncate(leftCutoff, rightCutoff, t))
DistributionTypes.PointSet(PointSetDist.T.truncate(leftCutoff, rightCutoff, t))
})
}
}
@ -168,7 +168,7 @@ module AlgebraicCombination = {
->E.R.bind(((t1, t2)) => {
SampleSetDist.map2(~fn, ~t1, ~t2)->GenericDist_Types.Error.resultStringToResultError
})
->E.R2.fmap(r => GenericDist_Types.SampleSet(r))
->E.R2.fmap(r => DistributionTypes.SampleSet(r))
}
//I'm (Ozzie) really just guessing here, very little idea what's best
@ -206,7 +206,7 @@ module AlgebraicCombination = {
arithmeticOperation,
t1,
t2,
)->E.R2.fmap(r => GenericDist_Types.PointSet(r))
)->E.R2.fmap(r => DistributionTypes.PointSet(r))
}
}
}
@ -229,7 +229,7 @@ let pointwiseCombination = (
t2,
)
)
->E.R2.fmap(r => GenericDist_Types.PointSet(r))
->E.R2.fmap(r => DistributionTypes.PointSet(r))
}
let pointwiseCombinationFloat = (
@ -239,7 +239,7 @@ let pointwiseCombinationFloat = (
~float: float,
): result<t, error> => {
let m = switch arithmeticOperation {
| #Add | #Subtract => Error(GenericDist_Types.DistributionVerticalShiftIsInvalid)
| #Add | #Subtract => Error(DistributionTypes.DistributionVerticalShiftIsInvalid)
| (#Multiply | #Divide | #Power | #Logarithm) as arithmeticOperation =>
toPointSetFn(t)->E.R2.fmap(t => {
//TODO: Move to PointSet codebase
@ -254,7 +254,7 @@ let pointwiseCombinationFloat = (
)
})
}
m->E.R2.fmap(r => GenericDist_Types.PointSet(r))
m->E.R2.fmap(r => DistributionTypes.PointSet(r))
}
//Note: The result should always cumulatively sum to 1. This would be good to test.
@ -265,7 +265,7 @@ let mixture = (
~pointwiseAddFn: pointwiseAddFn,
) => {
if E.A.length(values) == 0 {
Error(GenericDist_Types.Other("Mixture error: mixture must have at least 1 element"))
Error(DistributionTypes.Other("Mixture error: mixture must have at least 1 element"))
} else {
let totalWeight = values->E.A2.fmap(E.Tuple2.second)->E.A.Floats.sum
let properlyWeightedValues =

View File

@ -1,15 +1,6 @@
type genericDist =
| PointSet(PointSetTypes.pointSetDist)
| SampleSet(SampleSetDist.t)
| Symbolic(SymbolicDistTypes.symbolicDist)
type genericDist = DistributionTypes.genericDist
@genType
type error =
| NotYetImplemented
| Unreachable
| DistributionVerticalShiftIsInvalid
| ArgumentError(string)
| Other(string)
type error = DistributionTypes.error
@genType
module Error = {
@ -23,6 +14,7 @@ module Error = {
| NotYetImplemented => "Not Yet Implemented"
| Unreachable => "Unreachable"
| DistributionVerticalShiftIsInvalid => "Distribution Vertical Shift Is Invalid"
| ArgumentError(x) => `Argument Error: ${x}`
| Other(s) => s
}
}

View File

@ -141,6 +141,8 @@ module Lognormal = {
}
let divide = (l1, l2) => {
let mu = l1.mu -. l2.mu
// We believe the ratiands will have covariance zero.
// See here https://stats.stackexchange.com/questions/21735/what-are-the-mean-and-variance-of-the-ratio-of-two-lognormal-variables for details
let sigma = l1.sigma +. l2.sigma
#Lognormal({mu: mu, sigma: sigma})
}

View File

@ -11,10 +11,10 @@ The below few seem to work fine. In the future there's definitely more work to d
type samplingParams = DistributionOperation.env
@genType
type genericDist = GenericDist_Types.genericDist
type genericDist = DistributionTypes.genericDist
@genType
type distributionError = GenericDist_Types.error
type distributionError = DistributionTypes.error
@genType
type resultDist = result<genericDist, distributionError>

View File

@ -59,8 +59,9 @@ module O = {
let toExn = Rationale.Option.toExn
let some = Rationale.Option.some
let firstSome = Rationale.Option.firstSome
let toExt = Rationale.Option.toExn
let toExt = Rationale.Option.toExn // wanna flag this-- looks like a typo but `Rationale.OptiontoExt` doesn't exist.
let flatApply = (fn, b) => Rationale.Option.apply(fn, Some(b)) |> Rationale.Option.flatten
let flatten = Rationale.Option.flatten
let toBool = opt =>
switch opt {
@ -103,6 +104,7 @@ module O2 = {
let toExn = (a, b) => O.toExn(b, a)
let fmap = (a, b) => O.fmap(b, a)
let toResult = (a, b) => O.toResult(b, a)
let bind = (a, b) => O.bind(b, a)
}
/* Functions */
@ -176,6 +178,31 @@ module R = {
let errorIfCondition = (errorCondition, errorMessage, r) =>
errorCondition(r) ? Error(errorMessage) : Ok(r)
let ap = Rationale.Result.ap
let ap' = (r, a) =>
switch r {
| Ok(f) => fmap(f, a)
| Error(err) => Error(err)
}
// (a1 -> a2 -> r) -> m a1 -> m a2 -> m r // not in Rationale
let liftM2: (('a, 'b) => 'c, result<'a, 'd>, result<'b, 'd>) => result<'c, 'd> = (op, xR, yR) => {
ap'(fmap(op, xR), yR)
}
let liftJoin2: (('a, 'b) => result<'c, 'd>, result<'a, 'd>, result<'b, 'd>) => result<'c, 'd> = (
op,
xR,
yR,
) => {
bind(liftM2(op, xR, yR), x => x)
}
let fmap2 = (f, r) =>
switch r {
| Ok(r) => r->Ok
| Error(x) => x->f->Error
}
}
module R2 = {
@ -188,6 +215,12 @@ module R2 = {
| Ok(r) => Ok(r)
| Error(e) => map(e)
}
let fmap2 = (xR, f) =>
switch xR {
| Ok(x) => x->Ok
| Error(x) => x->f->Error
}
}
let safe_fn_of_string = (fn, s: string): option<'a> =>
@ -258,6 +291,29 @@ module L = {
let update = Rationale.RList.update
let iter = List.iter
let findIndex = Rationale.RList.findIndex
let headSafe = Belt.List.head
let tailSafe = Belt.List.tail
let headExn = Belt.List.headExn
let tailExn = Belt.List.tailExn
let zip = Belt.List.zip
let combinations2: list<'a> => list<('a, 'a)> = xs => {
let rec loop: ('a, list<'a>) => list<('a, 'a)> = (x', xs') => {
let n = length(xs')
if n == 0 {
list{}
} else {
let combs = fmap(y => (x', y), xs')
let hd = headExn(xs')
let tl = tailExn(xs')
concat(list{combs, loop(hd, tl)})
}
}
switch (headSafe(xs), tailSafe(xs)) {
| (Some(x'), Some(xs')) => loop(x', xs')
| (_, _) => list{}
}
}
}
/* A for Array */