linearly_separable.qmd 9.52 KiB
```{julia}
include("$(pwd())/notebooks/setup.jl")
eval(setup_notebooks)
```
# Linearly Separable Data
```{julia}
# Hyper:
_retrain = true
# Data:
test_size = 0.2
n_obs = Int(1000 / (1.0 - test_size))
counterfactual_data, test_data = train_test_split(
load_blobs(n_obs; cluster_std=0.1, center_box=(-1. => 1.));
test_size=test_size
)
X, y = CounterfactualExplanations.DataPreprocessing.unpack_data(counterfactual_data)
X = table(permutedims(X))
labels = counterfactual_data.output_encoder.labels
input_dim, n_obs = size(counterfactual_data.X)
output_dim = length(unique(labels))
```
First, let's create a couple of image classifier architectures:
```{julia}
# Model parameters:
epochs = 100
batch_size = minimum([Int(round(n_obs/10)), 128])
n_hidden = 16
activation = Flux.swish
builder = MLJFlux.MLP(
hidden=(n_hidden, n_hidden, n_hidden),
σ=Flux.swish
)
n_ens = 5 # number of models in ensemble
_loss = Flux.Losses.crossentropy # loss function
_finaliser = Flux.softmax # finaliser function
```
```{julia}
# JEM parameters:
𝒟x = Normal()
𝒟y = Categorical(ones(output_dim) ./ output_dim)
sampler = ConditionalSampler(
𝒟x, 𝒟y,
input_size=(input_dim,),
batch_size=50,
)
α = [1.0,1.0,1e-1] # penalty strengths
```
```{julia}
# Simple MLP:
mlp = NeuralNetworkClassifier(
builder=builder,
epochs=epochs,
batch_size=batch_size,
finaliser=_finaliser,
loss=_loss,
)
# Deep Ensemble:
mlp_ens = EnsembleModel(model=mlp, n=n_ens)
# Joint Energy Model:
jem = JointEnergyClassifier(
sampler; builder=builder,
epochs=epochs,
batch_size=batch_size,
finaliser=_finaliser,
loss=_loss,
jem_training_params=(
α=α,verbosity=10,
),
sampling_steps=30,
)
# JEM with adversarial training:
jem_adv = deepcopy(jem)
# jem_adv.adv_training = true
# Deep Ensemble of Joint Energy Models:
jem_ens = EnsembleModel(model=jem, n=n_ens)
# Deep Ensemble of Joint Energy Models with adversarial training:
# jem_ens_plus = EnsembleModel(model=jem_adv, n=n_ens)
# Dictionary of models:
models = Dict(
"MLP" => mlp,
# "MLP Ensemble" => mlp_ens,
"JEM" => jem,
# "JEM Ensemble" => jem_ens,
# "JEM Ensemble+" => jem_ens_plus,
)
```
```{julia}
# Train models:
function _train(model, X=X, y=labels; cov=0.95, method=:simple_inductive, mod_name="model")
conf_model = conformal_model(model; method=method, coverage=cov)
mach = machine(conf_model, X, y)
@info "Begin training $mod_name."
fit!(mach)
@info "Finished training $mod_name."
M = ECCCo.ConformalModel(mach.model, mach.fitresult)
return M
end
if _retrain
model_dict = Dict(mod_name => _train(model; mod_name=mod_name) for (mod_name, model) in models)
Serialization.serialize(joinpath(output_path,"linearly_separable_models.jls"), model_dict)
else
model_dict = Serialization.deserialize(joinpath(output_path,"linearly_separable_models.jls"))
end
```
```{julia}
# Evaluate models:
measure = Dict(
:f1score => multiclass_f1score,
:acc => accuracy,
:precision => multiclass_precision
)
model_performance = DataFrame()
for (mod_name, model) in model_dict
# Test performance:
_perf = CounterfactualExplanations.Models.model_evaluation(model, test_data, measure=collect(values(measure)))
_perf = DataFrame([[p] for p in _perf], collect(keys(measure)))
_perf.mod_name .= mod_name
model_performance = vcat(model_performance, _perf)
end
Serialization.serialize(joinpath(output_path,"linearly_separable_model_performance.jls"), model_performance)
CSV.write(joinpath(output_path, "linearly_separable_model_performance.csv"), model_performance)
model_performance
```
```{julia}
n_regen = 1000
n_each = batch_size
for (mod_name, model) in model_dict
K = length(counterfactual_data.y_levels)
input_size = size(selectdim(counterfactual_data.X, ndims(counterfactual_data.X), 1))
𝒟x = Uniform(extrema(counterfactual_data.X)...)
𝒟y = Categorical(ones(K) ./ K)
sampler = ConditionalSampler(𝒟x, 𝒟y; input_size=input_size)
opt = ImproperSGLD()
plts = []
for target in levels(labels)
target_idx = findall(levels(labels) .== target)[1]
f(x) = logits(model, x)
X̂ = sampler(f, opt; niter=n_regen, n_samples=n_each, y=target_idx)
ex = extrema(hcat(MLJFlux.reformat(X),X̂), dims=2)
xlims = ex[1]
ylims = ex[2]
x1 = range(1.0f0.*xlims...,length=10)
x2 = range(1.0f0.*ylims...,length=10)
p(x) = probs(model, x)
plt = Plots.contour(
x1, x2, (x, y) -> p([x, y][:,:])[target_idx],
fill=true, alpha=0.5, title="Target: $target", cbar=true,
xlims=xlims,
ylims=ylims,
)
Plots.scatter!(
MLJFlux.reformat(X)[1,:], MLJFlux.reformat(X)[2,:],
color=Int.(labels.refs), group=Int.(labels.refs), alpha=0.5
)
Plots.scatter!(
X̂[1,:], X̂[2,:],
color=repeat([target_idx], size(X̂,2)),
group=repeat([target_idx], size(X̂,2)),
shape=:star5, ms=10
)
savefig(plt, joinpath(output_images_path, "linearly_separable_generated_$(mod_name).png"))
push!(plts, plt)
end
plt = Plots.plot(plts..., layout=(1, 2), size=(2*500, 400), plot_title=mod_name)
display(plt)
end
```
```{julia}
#| output: true
#| echo: false
#| label: fig-losses
#| fig-cap: "Illustration of the smooth size loss and the configurable classification loss."
X_plot = matrix(X)
temp = 0.1
for (mod_name, model) in model_dict
p0 = Plots.contourf(model.model, model.fitresult, X_plot, labels; plot_set_size=true, zoom=0, temp=temp)
p1 = Plots.contourf(model.model, model.fitresult, X_plot, labels; plot_set_loss=true, zoom=0, temp=temp)
p2 = Plots.contourf(model.model, model.fitresult, X_plot, labels; plot_classification_loss=true, zoom=0, temp=temp, clim=nothing, loss_matrix=ones(2,2))
display(Plots.plot(p0, p1, p2, size=(1400,320), plot_title=mod_name, layout=(1,3)))
end
```
## Benchmark
```{julia}
λ₁ = 0.25
λ₂ = 0.75
λ₃ = 0.75
Λ = [λ₁, λ₂, λ₃]
opt = Flux.Optimise.Descent(0.01)
use_class_loss = false
# Benchmark generators:
generator_dict = Dict(
"Wachter" => WachterGenerator(λ=λ₁, opt=opt),
"REVISE" => REVISEGenerator(λ=λ₁, opt=opt),
"Schut" => GreedyGenerator(),
"ECCCo" => ECCCoGenerator(λ=Λ, opt=opt, use_class_loss=use_class_loss),
"ECCCo (no CP)" => ECCCoGenerator(λ=[λ₁, 0.0, λ₃], opt=opt, use_class_loss=use_class_loss),
"ECCCo (no EBM)" => ECCCoGenerator(λ=[λ₁, λ₂, 0.0], opt=opt, use_class_loss=use_class_loss),
)
```
### POC
```{julia}
Random.seed!(2023)
M = model_dict["JEM"]
X = X isa Matrix ? X : Float32.(permutedims(matrix(X)))
factual_label = levels(labels)[1]
x_factual = reshape(X[:,rand(findall(predict_label(M, counterfactual_data).==factual_label))],input_dim,1)
target = levels(labels)[2]
factual = predict_label(M, counterfactual_data, x_factual)[1]
ces = Dict{Any,Any}()
plts = []
for (name, generator) in generator_dict
ce = generate_counterfactual(
x_factual, target, counterfactual_data, M, generator;
initialization=:identity,
converge_when=:generator_conditions,
)
plt = Plots.plot(
ce, title=name, alpha=0.2,
cbar=false,
# axis=nothing,
)
if contains(name, "ECCCo")
_X = distance_from_energy(ce, return_conditionals=true)
Plots.scatter!(
_X[1,:],_X[2,:], color=:purple, shape=:star5,
ms=10, label="x̂|$target", alpha=0.5
)
end
push!(plts, plt)
ces[name] = ce
end
plt = Plots.plot(plts..., size=(650,500))
display(plt)
savefig(plt, joinpath(output_images_path, "linearly_separable_poc.png"))
```
### Complete Benchmark
```{julia}
# Measures:
measures = [
CounterfactualExplanations.distance,
ECCCo.distance_from_energy,
ECCCo.distance_from_targets,
CounterfactualExplanations.Evaluation.validity,
CounterfactualExplanations.Evaluation.redundancy,
ECCCo.set_size_penalty
]
bmks = []
for target in sort(unique(labels)) for factual in sort(unique(labels))
if factual == target
continue
end
bmk = benchmark(
counterfactual_data;
models=model_dict,
generators=generator_dict,
measure=measures,
suppress_training=true, dataname="Linearly Separable",
n_individuals=25,
target=target, factual=factual,
initialization=:identity,
converge_when=:generator_conditions,
)
push!(bmks, bmk)
end
end
bmk = reduce(vcat, bmks)
CSV.write(joinpath(output_path, "linearly_separable_benchmark.csv"), bmk())
```
```{julia}
df = @chain bmk() begin
@mutate(variable = ifelse.(variable .== "distance_from_energy", "Non-Conformity", variable))
@mutate(variable = ifelse.(variable .== "distance_from_targets", "Implausibility", variable))
@mutate(variable = ifelse.(variable .== "distance", "Cost", variable))
@mutate(variable = ifelse.(variable .== "redundancy", "Redundancy", variable))
@mutate(variable = ifelse.(variable .== "Validity", "Validity", variable))
end
plt = AlgebraOfGraphics.data(df) * visual(BoxPlot) *
mapping(:generator, :value, row=:variable, col=:model, color=:generator)
plt = draw(
plt, axis=(xlabel="", xticksvisible=false, xticklabelsvisible=false, width=150, height=120),
facet=(; linkyaxes=:none)
)
display(plt)
save(joinpath(output_images_path, "linearly_separable_benchmark.png"), plt, px_per_unit=5)
```