Linear regression
using DynamicHMCModels
ProjDir = rel_path_d("..", "scripts", "05")
cd(ProjDir)
Import the dataset.
snippet 5.4
wd = CSV.read(rel_path("..", "data", "WaffleDivorce.csv"), delim=';')
df = convert(DataFrame, wd);
mean_ma = mean(df[!, :Marriage])
df[!, :Marriage_s] = convert(Vector{Float64},
(df[!, :Marriage]) .- mean_ma)/std(df[!, :Marriage]);
mean_mam = mean(df[!, :MedianAgeMarriage])
df[!, :MedianAgeMarriage_s] = convert(Vector{Float64},
(df[!, :MedianAgeMarriage]) .- mean_mam)/std(df[!, :MedianAgeMarriage]);
50-element Array{Float64,1}:
-0.6062895051354262
-0.6866992538271283
-0.20424076167692148
-1.4103869920524357
0.5998567252400879
-0.28465051036862354
1.2431347147736962
0.43903722785668664
2.9317394372994143
0.27821773047328247
⋮
-0.6866992538271283
-0.6866992538271283
-2.214484478969445
0.6802664739317872
0.27821773047328247
-0.12383101298522224
-0.8475187512105296
0.19780798178158324
-1.4907967407441378
Show the first six rows of the dataset.
first(df[!, [1, 7, 14,15]], 6)
Location | Divorce | Marriage_s | MedianAgeMarriage_s | |
---|---|---|---|---|
String | Float64 | Float64 | Float64 | |
1 | Alabama | 12.7 | 0.0226441 | -0.60629 |
2 | Alaska | 12.5 | 1.5498 | -0.686699 |
3 | Arizona | 10.8 | 0.0489744 | -0.204241 |
4 | Arkansas | 13.5 | 1.65512 | -1.41039 |
5 | California | 8.0 | -0.266989 | 0.599857 |
6 | Colorado | 11.6 | 0.891544 | -0.284651 |
Model $y ∼ Xβ + ϵ$, where $ϵ ∼ N(0, σ²)$ IID. Student prior on σ
struct m_5_3{TY <: AbstractVector, TX <: AbstractMatrix}
"Observations."
y::TY
"Covariates"
X::TX
end
Make the type callable with the parameters as a single argument.
function (problem::m_5_3)(θ)
@unpack y, X, = problem # extract the data
@unpack β, σ = θ # works on the named tuple too
ll = 0.0
ll += logpdf(Normal(10, 10), X[1]) # a = X[1]
ll += logpdf(Normal(0, 1), X[2]) # b1 = X[2]
ll += logpdf(Normal(0, 1), X[3]) # b1 = X[3]
ll += logpdf(TDist(1.0), σ)
ll += loglikelihood(Normal(0, σ), y .- X*β)
ll
end
Instantiate the model with data and inits.
N = size(df, 1)
X = hcat(ones(N), df[!, :Marriage_s], df[!, :MedianAgeMarriage_s]);
y = convert(Vector{Float64}, df[!, :Divorce])
p = m_5_3(y, X);
p((β = [1.0, 2.0, 3.0], σ = 1.0))
-2222.175273500088
Write a function to return properly dimensioned transformation.
problem_transformation(p::m_5_3) =
as((β = as(Array, size(p.X, 2)), σ = asℝ₊))
problem_transformation (generic function with 1 method)
Wrap the problem with a transformation, then use Flux for the gradient.
P = TransformedLogDensity(problem_transformation(p), p)
∇P = ADgradient(:ForwardDiff, P);
ForwardDiff AD wrapper for TransformedLogDensity of dimension 4, w/ chunk size 4
Tune and sample.
chain, NUTS_tuned = NUTS_init_tune_mcmc(∇P, 1000);
(NUTS_Transition{Array{Float64,1},Float64}[NUTS_Transition{Array{Float64,1},Float64}([9.44859839780487, -0.38822720992007287, -1.2063909833808442, 0.3060722742054358], -101.5016375107753, 2, DynamicHMC.AdjacentTurn, 0.9555681429989432, 7), NUTS_Transition{Array{Float64,1},Float64}([9.477748270509286, -0.22179098622197196, -0.9739309388564404, 0.33493555952395154], -100.18687471355072, 3, DynamicHMC.DoubledTurn, 0.9676507339397684, 7), NUTS_Transition{Array{Float64,1},Float64}([9.660700097092928, -0.25905977771917593, -1.3664411707177266, 0.4699387900407548], -99.34451345923559, 2, DynamicHMC.DoubledTurn, 0.9858957869169793, 3), NUTS_Transition{Array{Float64,1},Float64}([9.901473245396385, -0.685280558297139, -1.4309784425536518, 0.5020299483898931], -102.54894318505188, 1, DynamicHMC.AdjacentTurn, 0.7153640173271075, 3), NUTS_Transition{Array{Float64,1},Float64}([9.611805549501947, 0.10058266887514294, -1.3762127315396824, 0.39397522770119164], -101.86775403249462, 2, DynamicHMC.AdjacentTurn, 0.9194731679246563, 7), NUTS_Transition{Array{Float64,1},Float64}([9.632009348797693, -0.3597538304948612, -1.4440707920872293, 0.4504951157595679], -100.98014720078153, 2, DynamicHMC.DoubledTurn, 0.9400394114929629, 3), NUTS_Transition{Array{Float64,1},Float64}([9.497904993523461, -0.18382762383042134, -1.3608648272096973, 0.45746120028442605], -98.94934097766524, 2, DynamicHMC.AdjacentTurn, 0.9723065665259837, 7), NUTS_Transition{Array{Float64,1},Float64}([9.670926433644762, -0.45583801001007873, -1.5494999616307734, 0.30467701550787524], -99.17101173474788, 3, DynamicHMC.DoubledTurn, 0.9941309802823081, 7), NUTS_Transition{Array{Float64,1},Float64}([9.563095152771764, -0.13878242450858486, -1.0637828459963596, 0.3072849999686285], -100.41750307928406, 3, DynamicHMC.AdjacentTurn, 0.856852782875545, 15), NUTS_Transition{Array{Float64,1},Float64}([9.651755118014494, -0.04547826039031169, -0.888785739815467, 0.38380733621986746], -99.04324008323319, 2, DynamicHMC.AdjacentTurn, 0.960638503352037, 7) … NUTS_Transition{Array{Float64,1},Float64}([9.72871302863776, -0.5036983606124653, -1.8000975674115414, 0.37581294074594385], -99.96374627796848, 2, DynamicHMC.DoubledTurn, 0.8672844572135731, 3), NUTS_Transition{Array{Float64,1},Float64}([9.918965677864591, 0.031625860465330086, -0.6479073777168369, 0.4043617636718365], -101.05692212257783, 3, DynamicHMC.DoubledTurn, 0.9516132425613971, 7), NUTS_Transition{Array{Float64,1},Float64}([9.6211505124098, -0.4107979399844264, -1.5032186924926172, 0.22701626919251125], -104.44322872876324, 3, DynamicHMC.DoubledTurn, 0.7251810126352941, 7), NUTS_Transition{Array{Float64,1},Float64}([9.652591702022093, -0.2953741042051475, -1.1910488192167263, 0.3435665847421631], -100.07703922944597, 2, DynamicHMC.DoubledTurn, 0.9052906834599247, 3), NUTS_Transition{Array{Float64,1},Float64}([9.786421450345038, -0.1750790955522674, -1.3037080220365704, 0.41489081978325476], -99.02040830711982, 3, DynamicHMC.DoubledTurn, 0.9268681226228843, 7), NUTS_Transition{Array{Float64,1},Float64}([9.638181409247348, -0.14547320571315575, -1.4723258729910864, 0.3475958490675849], -100.4741619593947, 2, DynamicHMC.DoubledTurn, 0.815649776492292, 3), NUTS_Transition{Array{Float64,1},Float64}([9.8656736339535, -0.6513476579805358, -1.1556781639097735, 0.4921719023871601], -101.34601581611095, 2, DynamicHMC.AdjacentTurn, 0.7762830561593663, 7), NUTS_Transition{Array{Float64,1},Float64}([9.569871220959001, 0.35133403015349357, -1.144362105838841, 0.347622862857632], -101.96024528828711, 3, DynamicHMC.DoubledTurn, 0.9800892748238047, 7), NUTS_Transition{Array{Float64,1},Float64}([9.64317820093146, -0.6805983828949609, -1.586702739210891, 0.2333747775486135], -103.08230011011821, 3, DynamicHMC.DoubledTurn, 0.9865568480406027, 7), NUTS_Transition{Array{Float64,1},Float64}([9.565589171827192, 0.1969432457210854, -0.9640338591631108, 0.5028715467905426], -100.3336189876059, 3, DynamicHMC.DoubledTurn, 1.0, 7)], NUTS sampler in 4 dimensions
stepsize (ϵ) ≈ 0.75
maximum depth = 10
Gaussian kinetic energy, √diag(M⁻¹): [0.19281852893239879, 0.3063603664098533, 0.3075931116578144, 0.1021851290592024]
)
We use the transformation to obtain the posterior from the chain.
posterior = TransformVariables.transform.(Ref(problem_transformation(p)), get_position.(chain));
posterior[1:5]
5-element Array{NamedTuple{(:β, :σ),Tuple{Array{Float64,1},Float64}},1}:
(β = [9.44859839780487, -0.38822720992007287, -1.2063909833808442], σ = 1.3580804571869312)
(β = [9.477748270509286, -0.22179098622197196, -0.9739309388564404], σ = 1.3978503041810197)
(β = [9.660700097092928, -0.25905977771917593, -1.3664411707177266], σ = 1.599896260635256)
(β = [9.901473245396385, -0.685280558297139, -1.4309784425536518], σ = 1.65207148902368)
(β = [9.611805549501947, 0.10058266887514294, -1.3762127315396824], σ = 1.4828638142742578)
Extract the parameter posterior means: β
,
posterior_β = mean(first, posterior)
3-element Array{Float64,1}:
9.675756289333941
-0.2170463173406174
-1.2351324361764504
then σ
:
posterior_σ = mean(last, posterior)
1.5031993651540325
Effective sample sizes (of untransformed draws)
ess = mapslices(effective_sample_size,
get_position_matrix(chain); dims = 1)
1×4 Array{Float64,2}:
864.581 904.233 875.131 886.911
NUTS-specific statistics
NUTS_statistics(chain)
Hamiltonian Monte Carlo sample of length 1000
acceptance rate mean: 0.91, min/25%/median/75%/max: 0.44 0.86 0.95 0.99 1.0
termination: AdjacentTurn => 16% DoubledTurn => 84%
depth: 1 => 2% 2 => 47% 3 => 50% 4 => 1% 5 => 0%
cmdstan result
cmdstan_result = "
Iterations = 1:1000
Thinning interval = 1
Chains = 1,2,3,4
Samples per chain = 1000
Empirical Posterior Estimates:
Mean SD Naive SE MCSE ESS
a 9.69137275 0.21507432 0.0034006235 0.0038501180 1000
bA -1.12184710 0.29039965 0.0045916216 0.0053055477 1000
bM -0.12106472 0.28705400 0.0045387223 0.0051444688 1000
sigma 1.52326545 0.16272599 0.0025729239 0.0034436330 1000
Quantiles:
2.5% 25.0% 50.0% 75.0% 97.5%
a 9.2694878 9.5497650 9.6906850 9.83227750 10.11643500
bA -1.6852295 -1.3167700 -1.1254650 -0.92889225 -0.53389157
bM -0.6889247 -0.3151695 -0.1231065 0.07218513 0.45527243
sigma 1.2421182 1.4125950 1.5107700 1.61579000 1.89891925
";
"\nIterations = 1:1000\nThinning interval = 1\nChains = 1,2,3,4\nSamples per chain = 1000\n\nEmpirical Posterior Estimates:\n Mean SD Naive SE MCSE ESS\n a 9.69137275 0.21507432 0.0034006235 0.0038501180 1000\n bA -1.12184710 0.29039965 0.0045916216 0.0053055477 1000\n bM -0.12106472 0.28705400 0.0045387223 0.0051444688 1000\nsigma 1.52326545 0.16272599 0.0025729239 0.0034436330 1000\n\nQuantiles:\n 2.5% 25.0% 50.0% 75.0% 97.5%\n a 9.2694878 9.5497650 9.6906850 9.83227750 10.11643500\n bA -1.6852295 -1.3167700 -1.1254650 -0.92889225 -0.53389157\n bM -0.6889247 -0.3151695 -0.1231065 0.07218513 0.45527243\nsigma 1.2421182 1.4125950 1.5107700 1.61579000 1.89891925\n"
Extract the parameter posterior means: [β, σ]
,
[posterior_β, posterior_σ]
2-element Array{Any,1}:
[9.675756289333941, -0.2170463173406174, -1.2351324361764504]
1.5031993651540325
end of m4.5d.jl#- This page was generated using Literate.jl.