using Turing
using FillArrays
using Lux
using Plots
import Mooncake
using Functors
using LinearAlgebra
using Random
Bayesian Neural Networks
In this tutorial, we demonstrate how one can implement a Bayesian Neural Network using a combination of Turing and Lux, a suite of machine learning tools. We will use Lux to specify the neural network’s layers and Turing to implement the probabilistic inference, with the goal of implementing a classification algorithm.
We will begin with importing the relevant libraries.
Our goal here is to use a Bayesian neural network to classify points in an artificial dataset. The code below generates data points arranged in a box-like pattern and displays a graph of the dataset we will be working with.
# Number of points to generate
= 80
N = round(Int, N / 4)
M = Random.default_rng()
rng Random.seed!(rng, 1234)
# Generate artificial data
= rand(rng, Float32, M) * 4.5f0;
x1s = rand(rng, Float32, M) * 4.5f0;
x2s = Array([[x1s[i] + 0.5f0; x2s[i] + 0.5f0] for i in 1:M])
xt1s = rand(rng, Float32, M) * 4.5f0;
x1s = rand(rng, Float32, M) * 4.5f0;
x2s append!(xt1s, Array([[x1s[i] - 5.0f0; x2s[i] - 5.0f0] for i in 1:M]))
= rand(rng, Float32, M) * 4.5f0;
x1s = rand(rng, Float32, M) * 4.5f0;
x2s = Array([[x1s[i] + 0.5f0; x2s[i] - 5.0f0] for i in 1:M])
xt0s = rand(rng, Float32, M) * 4.5f0;
x1s = rand(rng, Float32, M) * 4.5f0;
x2s append!(xt0s, Array([[x1s[i] - 5.0f0; x2s[i] + 0.5f0] for i in 1:M]))
# Store all the data for later
= [xt1s; xt0s]
xs = [ones(2 * M); zeros(2 * M)]
ts
# Plot data points.
function plot_data()
= map(e -> e[1], xt1s)
x1 = map(e -> e[2], xt1s)
y1 = map(e -> e[1], xt0s)
x2 = map(e -> e[2], xt0s)
y2
scatter(x1, y1; color="red", clim=(0, 1))
Plots.return Plots.scatter!(x2, y2; color="blue", clim=(0, 1))
end
plot_data()
Building a Neural Network
The next step is to define a feedforward neural network where we express our parameters as distributions, and not single points as with traditional neural networks. For this we will use Dense
to define liner layers and compose them via Chain
, both are neural network primitives from Lux. The network nn_initial
we created has two hidden layers with tanh
activations and one output layer with sigmoid (σ
) activation, as shown below.
The nn_initial
is an instance that acts as a function and can take data as inputs and output predictions. We will define distributions on the neural network parameters.
# Construct a neural network using Lux
= Chain(Dense(2 => 3, tanh), Dense(3 => 2, tanh), Dense(2 => 1, σ))
nn_initial
# Initialize the model weights and state
= Lux.setup(rng, nn_initial)
ps, st
parameterlength(nn_initial) # number of parameters in NN Lux.
20
The probabilistic model specification below creates a parameters
variable, which has IID normal variables. The parameters
vector represents all parameters of our neural net (weights and biases).
# Create a regularization term and a Gaussian prior variance term.
= 0.09
alpha = sqrt(1.0 / alpha) sigma
3.3333333333333335
We also define a function to construct a named tuple from a vector of sampled parameters. (We could use ComponentArrays
here and broadcast to avoid doing this, but this way avoids introducing an extra dependency.)
function vector_to_parameters(ps_new::AbstractVector, ps::NamedTuple)
@assert length(ps_new) == Lux.parameterlength(ps)
= 1
i function get_ps(x)
= reshape(view(ps_new, i:(i + length(x) - 1)), size(x))
z += length(x)
i return z
end
return fmap(get_ps, ps)
end
vector_to_parameters (generic function with 1 method)
To interface with external libraries it is often desirable to use the StatefulLuxLayer
to automatically handle the neural network states.
const nn = StatefulLuxLayer{true}(nn_initial, nothing, st)
# Specify the probabilistic model.
@model function bayes_nn(xs, ts; sigma = sigma, ps = ps, nn = nn)
# Sample the parameters
= Lux.parameterlength(nn_initial)
nparameters ~ MvNormal(zeros(nparameters), Diagonal(abs2.(sigma .* ones(nparameters))))
parameters
# Forward NN to make predictions
= Lux.apply(nn, xs, f32(vector_to_parameters(parameters, ps)))
preds
# Observe each prediction.
for i in eachindex(ts)
~ Bernoulli(preds[i])
ts[i] end
end
bayes_nn (generic function with 2 methods)
Inference can now be performed by calling sample
. We use the NUTS
Hamiltonian Monte Carlo sampler here.
setprogress!(false)
# Perform inference.
= 2_000
n_iters = sample(bayes_nn(reduce(hcat, xs), ts), NUTS(; adtype=AutoMooncake(; config=nothing)), n_iters); ch
┌ Info: Found initial step size
└ ϵ = 0.4
Now we extract the parameter samples from the sampled chain as θ
(this is of size 5000 x 20
where 5000
is the number of iterations and 20
is the number of parameters). We’ll use these primarily to determine how good our model’s classifier is.
# Extract all weight and bias parameters.
= MCMCChains.group(ch, :parameters).value; θ
Prediction Visualization
We can use MAP estimation to classify our population by using the set of weights that provided the highest log posterior.
# A helper to run the nn through data `x` using parameters `θ`
nn_forward(x, θ) = nn(x, vector_to_parameters(θ, ps))
# Plot the data we have.
= plot_data()
fig
# Find the index that provided the highest log posterior in the chain.
= findmax(ch[:lp])
_, i
# Extract the max row value from i.
= i.I[1]
i
# Plot the posterior distribution with a contour plot
= collect(range(-6; stop=6, length=25))
x1_range = collect(range(-6; stop=6, length=25))
x2_range = [nn_forward([x1, x2], θ[i, :])[1] for x1 in x1_range, x2 in x2_range]
Z contour!(x1_range, x2_range, Z; linewidth=3, colormap=:seaborn_bright)
fig
The contour plot above shows that the MAP method is not too bad at classifying our data.
Now we can visualize our predictions.
\[ p(\tilde{x} | X, \alpha) = \int_{\theta} p(\tilde{x} | \theta) p(\theta | X, \alpha) \approx \sum_{\theta \sim p(\theta | X, \alpha)}f_{\theta}(\tilde{x}) \]
The nn_predict
function takes the average predicted value from a network parameterized by weights drawn from the MCMC chain.
# Return the average predicted value across
# multiple weights.
function nn_predict(x, θ, num)
= min(num, size(θ, 1)) # make sure num does not exceed the number of samples
num return mean([first(nn_forward(x, view(θ, i, :))) for i in 1:10:num])
end
nn_predict (generic function with 1 method)
Next, we use the nn_predict
function to predict the value at a sample of points where the x1
and x2
coordinates range between -6 and 6. As we can see below, we still have a satisfactory fit to our data, and more importantly, we can also see where the neural network is uncertain about its predictions much easier—those regions between cluster boundaries.
# Plot the average prediction.
= plot_data()
fig
= 1500
n_end = collect(range(-6; stop=6, length=25))
x1_range = collect(range(-6; stop=6, length=25))
x2_range = [nn_predict([x1, x2], θ, n_end)[1] for x1 in x1_range, x2 in x2_range]
Z contour!(x1_range, x2_range, Z; linewidth=3, colormap=:seaborn_bright)
fig
Suppose we are interested in how the predictive power of our Bayesian neural network evolved between samples. In that case, the following graph displays an animation of the contour plot generated from the network weights in samples 1 to 1,000.
# Number of iterations to plot.
= 500
n_end
= @gif for i in 1:n_end
anim plot_data()
= [nn_forward([x1, x2], θ[i, :])[1] for x1 in x1_range, x2 in x2_range]
Z contour!(x1_range, x2_range, Z; title="Iteration $i", clim=(0, 1))
end every 5
[ Info: Saved animation to /tmp/jl_45qDKX6dm7.gif
This has been an introduction to the applications of Turing and Lux in defining Bayesian neural networks.