Credal Wrapper on MNIST

The Credal Wrapper builds an ensemble of independently trained classifiers and represents the predictions as a Probability Intervals Credal Set. Epistemic uncertainty is captured by the width of the probability intervals: wide intervals indicate disagreement between ensemble members.

from __future__ import annotations

import numpy as np
import torch
from torch import nn

from probly.method.credal_wrapper import credal_wrapper
from probly.quantification import quantify
from probly.representer import representer
from probly_benchmark.data import load_mnist

from examples.utils.model import MLPClassifier
from examples.utils.plotting import plot_mnist_uncertainty

Setup

train_loader, test_loader = load_mnist(batch_size=256)

X_test_batches, y_test_batches = zip(*test_loader)
X_test = torch.cat([x.view(-1, 28 * 28) for x in X_test_batches])
y_test = torch.cat(list(y_test_batches))
images_test = (X_test.view(-1, 28, 28) * 255).byte()

Model

Each member is a softmax classifier; credal_wrapper aggregates their predictions into probability intervals.

base_model = MLPClassifier(in_features=28 * 28, hidden_features=256, out_features=10)
credal_model = credal_wrapper(
    base_model,
    predictor_type="logit_classifier",
    num_members=5,
)

Training

Each member is trained independently from a fresh initialization to maximize prediction diversity.

for member in credal_model:
    member.train()
    opt = torch.optim.Adam(member.parameters(), lr=1e-3)
    for _epoch in range(5):
        correct, total = 0, 0
        for X_batch, y_batch in train_loader:
            X_flat = X_batch.view(-1, 28 * 28)
            opt.zero_grad()
            logits = member(X_flat)
            loss = nn.functional.cross_entropy(logits, y_batch)
            loss.backward()
            opt.step()
            correct += (logits.detach().argmax(-1) == y_batch).sum().item()
            total += len(y_batch)
        if correct / total >= 0.97:
            break
    member.eval()

Uncertainty Quantification

The representer builds a credal set from all member predictions. quantify returns a CredalSetEntropyDecomposition where total uncertainty is the upper entropy and epistemic is the entropy gap.

rep = representer(credal_model)

with torch.no_grad():
    credal_set = rep.represent(X_test)

uq = quantify(credal_set)
_unc = uq.total if hasattr(uq, "total") else (uq.epistemic if hasattr(uq, "epistemic") else uq.aleatoric)
uncertainty = _unc.detach().numpy() if isinstance(_unc, torch.Tensor) else np.asarray(_unc)
uncertainty = uncertainty / np.log(2)
if uncertainty.ndim > 1:
    uncertainty = uncertainty.sum(axis=-1)

Predictions

with torch.no_grad():
    member_probs = torch.stack([member(X_test).softmax(-1) for member in credal_model]).numpy()
mean_probs = member_probs.mean(0)

accuracy = (mean_probs.argmax(-1) == y_test.numpy()).mean() * 100
print(f"Test accuracy: {accuracy:.1f}%")

Test accuracy: 97.7%

Visualization

plot = plot_mnist_uncertainty(
    images_test,
    y_test,
    uncertainty,
    mean_probs,
    title="Top-5 Most Uncertain Test Predictions (Credal Wrapper)",
)
plot.show()