torch_ecg.utils.top_n_accuracy¶

torch_ecg.utils.top_n_accuracy(labels: ndarray | Tensor, outputs: ndarray | Tensor, n: int | Sequence[int] = 1) → float | Dict[str, float][source]¶

Compute top n accuracy.

Parameters:

labels (numpy.ndarray or torch.Tensor) – Labels of class indices, of shape (batch_size,) or (batch_size, d_1, ..., d_m).
outputs (numpy.ndarray or torch.Tensor) – Predicted probabilities, of shape (batch_size, num_classes) or (batch_size, d_1, ..., d_m, num_classes) or (batch_size, num_classes, d_1, ..., d_m).
n (int or List[int]) – Top n to be considered.

Returns:

acc – Top n accuracy.

Return type:

float or dict of float

Examples

>>> from torch_ecg.cfg import DEFAULTS
>>> labels, outputs = DEFAULTS.RNG_randint(0, 9, (100)), DEFAULTS.RNG.uniform(0, 1, (100, 10))  # 100 samples, 10 classes
>>> top_n_accuracy(labels, outputs, 3)
0.32
>>> top_n_accuracy(labels, outputs, [1,3,5])
{'top_1_acc': 0.12, 'top_3_acc': 0.32, 'top_5_acc': 0.52}