d2l-ai/d2l-zh · error · AssertionError

test_acc <= 1 and test_acc > 0.7

Error message

test_acc <= 1 and test_acc > 0.7

What it means

Python AssertionError from d2l.torch.train_ch3's final sanity check: test accuracy must be in (0.7, 1] after training on Fashion-MNIST. It verifies the trained model generalizes as the book claims (~0.83 test acc); failure means underfitting, divergence, or an evaluation-path bug rather than a library defect.

Source

Thrown at d2l/torch.py:341

            self.axes[0].plot(x, y, fmt)
        self.config_axes()
        display.display(self.fig)
        display.clear_output(wait=True)

def train_ch3(net, train_iter, test_iter, loss, num_epochs, updater):
    """训练模型（定义见第3章）

    Defined in :numref:`sec_softmax_scratch`"""
    animator = Animator(xlabel='epoch', xlim=[1, num_epochs], ylim=[0.3, 0.9],
                        legend=['train loss', 'train acc', 'test acc'])
    for epoch in range(num_epochs):
        train_metrics = train_epoch_ch3(net, train_iter, loss, updater)
        test_acc = evaluate_accuracy(net, test_iter)
        animator.add(epoch + 1, train_metrics + (test_acc,))
    train_loss, train_acc = train_metrics
    assert train_loss < 0.5, train_loss
    assert train_acc <= 1 and train_acc > 0.7, train_acc
    assert test_acc <= 1 and test_acc > 0.7, test_acc

def predict_ch3(net, test_iter, n=6):
    """预测标签（定义见第3章）

    Defined in :numref:`sec_softmax_scratch`"""
    for X, y in test_iter:
        break
    trues = d2l.get_fashion_mnist_labels(y)
    preds = d2l.get_fashion_mnist_labels(d2l.argmax(net(X), axis=1))
    titles = [true +'\n' + pred for true, pred in zip(trues, preds)]
    d2l.show_images(
        d2l.reshape(X[0:n], (n, 28, 28)), 1, n, titles=titles[0:n])

def evaluate_loss(net, data_iter, loss):
    """评估给定数据集上模型的损失

    Defined in :numref:`sec_model_selection`"""
    metric = d2l.Accumulator(2)  # 损失的总和,样本数量

View on GitHub (pinned to e6b18ccea7)

Solutions

1. Rerun with the book's configuration: fresh net, num_epochs=10, lr=0.1, batch_size=256
2. Build train_iter and test_iter with identical transforms (only shuffle=True vs False differs)
3. Confirm net.eval() semantics if your custom model has dropout/batchnorm (d2l's evaluate_accuracy handles the standard softmax net)
4. If train acc is high but test acc ~0.1, check label order/argmax axis and that the same net object is passed to evaluation

Example fix

# before
train_iter = load_data_fashion_mnist(batch_size, resize=None)[0]  # normalized
_, test_iter = load_data_fashion_mnist(batch_size)              # different path
train_ch3(net, train_iter, test_iter, loss, 2, updater)  # test_acc 0.5 -> AssertionError
# after
train_iter, test_iter = load_data_fashion_mnist(256)
net = nn.Sequential(nn.Flatten(), nn.Linear(784, 10))
trainer = torch.optim.SGD(net.parameters(), lr=0.1)
train_ch3(net, train_iter, test_iter, loss, 10, trainer.step)  # test_acc ~0.83

train_iter, test_iter = d2l.load_data_fashion_mnist(256)  # same transforms
net = build_fresh_net()  # rebuild, never reuse stale weights
assert next(iter(test_iter)) is not None

def eval_pipeline_ready(net, test_iter) -> bool:
    X, y = next(iter(test_iter))
    return net(X).shape[1] == 10 and X.shape[0] == y.shape[0]

try:
    train_ch3(net, train_iter, test_iter, loss, num_epochs, updater)
except AssertionError as e:
    print(f'test_acc={e.args[0]} outside (0.7, 1]; check epochs, transforms, net freshness')
    raise

When it happens

Trigger: evaluate_accuracy(net, test_iter) returning <= 0.7: model undertrained (num_epochs=1-3), diverged lr, net reused from a previous overfit state, or test_iter normalized differently from train_iter (e.g. ToTensor only on the test transform path by mistake); also evaluating with net still in train() mode so dropout/batchnorm skew results if a custom net uses them.

Common situations: CPU-only environments where users trim num_epochs; reusing a net variable across notebook cells without re-instantiation; transforms applied inconsistently between the two DataLoaders; running on torch>=2.6 where default DataLoader/factory settings changed and iterators need explicit handling.

Related errors

• test_acc <= 1 and test_acc > 0.7
• train_loss < 0.5
• train_loss < 0.5
• train_acc <= 1 and train_acc > 0.7
• f"{name} 不存在于 {DATA_HUB}"

AI-assisted analysis of d2l-ai/d2l-zh@e6b18ccea7 (2026-08-14). Data as JSON: /api/errors/fe78a917d06d2588. Report an issue: GitHub.