squiggle/packages/squiggle-lang/__tests__/Distributions/Dotwise_test.res
Quinn Dougherty 0baeedfb46 pointwiseSubtract test; logscale test
Value: [1e-5 to 5e-4]
2022-05-04 11:42:51 -04:00

51 lines
1.7 KiB
Plaintext

open Jest
open Expect
open TestHelpers
open FastCheck
open Arbitrary
open Property.Sync
describe("dotSubtract", () => {
test("mean of normal minus exponential (unit)", () => {
let mean = 0.0
let rate = 10.0
let dotDifference = DistributionOperation.Constructors.pointwiseSubtract(
~env,
mkNormal(mean, 1.0),
mkExponential(rate),
)
let meanResult = E.R2.bind(DistributionOperation.Constructors.mean(~env), dotDifference)
let meanAnalytical = mean -. 1.0 /. rate
switch meanResult {
| Ok(meanValue) => meanValue->expect->toBeCloseTo(meanAnalytical)
| Error(err) => err->expect->toBe(DistributionTypes.OperationError(DivisionByZeroError))
}
})
Skip.test("mean of normal minus exponential (property)", () => {
assert_(
property2(float_(), floatRange(1e-5, 1e5), (mean, rate) => {
// We limit ourselves to stdev=1 so that the integral is trivial
let dotDifference = DistributionOperation.Constructors.pointwiseSubtract(
~env,
mkNormal(mean, 1.0),
mkExponential(rate),
)
let meanResult = E.R2.bind(DistributionOperation.Constructors.mean(~env), dotDifference)
// according to algebra or random variables,
let meanAnalytical = mean -. 1.0 /. rate
Js.Console.log3(
mean,
rate,
E.R.fmap(x => abs_float(x -. meanAnalytical) /. abs_float(meanAnalytical), meanResult),
)
switch meanResult {
| Ok(meanValue) => abs_float(meanValue -. meanAnalytical) /. abs_float(meanValue) < 1e-2 // 1% relative error
| Error(err) => err === DistributionTypes.OperationError(DivisionByZeroError)
}
}),
)
pass
})
})