Created
January 2, 2023 03:29
-
-
Save dlakelan/6204895ecce9b91040745ed5662daff1 to your computer and use it in GitHub Desktop.
Path Integral Approach For Functional Priors
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
using Pkg | |
Pkg.activate(".") | |
Pkg.add(["Turing","Distributions","ApproxFun","QuadGK","StatsPlots","LinearAlgebra","DataFrames"]) | |
using Turing, Distributions, ApproxFun, QuadGK, StatsPlots, LinearAlgebra, DataFrames | |
# a prior over functions that go up then come down... First let's see how ApproxFun works | |
Random.set_global_seed!(123456) | |
f = Fun(Chebyshev(Interval(0..100)),randn(12)) | |
plot(collect(0:100),map(f,0:100)) | |
@model function updown(n) | |
c ~ MvNormal(repeat([0.0],n),1000.0^2*I(n)) | |
## candidate functions rise from 0 to 1 over 0-20, stay near constant out to 50, | |
## and decline exponentially ish down to 0.3 out to 100 | |
## this could be some kind of function for say immune response as a function | |
## of age in years, or whatever | |
f = Fun(Chebyshev(Interval(0..100)),c) | |
a = quadgk(x -> ((f(x) - x/20)/.25)^2,0,20) # near linear rise over x 0..20 | |
Turing.@addlogprob!(-a[1]) | |
b = quadgk(x -> ((f(x) - 1.0)/.1)^2,20,50) # near 1 for x in 20-50 | |
Turing.@addlogprob!(-b[1]) | |
c = quadgk(x -> (((f(x) - (exp(-(x-50)/10)+0.2))/.25)^2),50,100) #exponential decay to 0.2 | |
Turing.@addlogprob!(-c[1]) | |
## relatively smooth functions | |
ddf = Derivative(2)*f | |
d = quadgk(x -> (ddf(x)^2),0,100) | |
Turing.@addlogprob!(-d[1]/100.0/.2) | |
end | |
s = Turing.sample(updown(10),NUTS(300,.8),300) | |
df = DataFrame(s) | |
p = plot(;legend=false) | |
x = collect(0.0:100.0) | |
for i in 1:10:300 | |
plot!(x,map(Fun(Chebyshev(Interval(0..100)),Vector(df[i,3:12])),x)) | |
end | |
display(p) |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment