speechbrain.utils.Accuracy module

Calculate accuracy.

Authors * Jianyuan Zhong 2020

Summary

Classes:

AccuracyStats

Module for calculate the overall one-step-forward prediction accuracy.

Functions:

Accuracy

Calculates the accuracy for predicted log probabilities and targets in a batch.

Reference

speechbrain.utils.Accuracy.Accuracy(log_probabilities, targets, length=None)

Calculates the accuracy for predicted log probabilities and targets in a batch.

Parameters:

• log_probabilities (torch.Tensor) – Predicted log probabilities (batch_size, time, feature).

• targets (torch.Tensor) – Target (batch_size, time).

• length (torch.Tensor) – Length of target (batch_size,).

Returns:

• numerator (float) – The number of correct samples

• denominator (float) – The total number of samples

Example

>>> probs = torch.tensor([[0.9, 0.1], [0.1, 0.9], [0.8, 0.2]]).unsqueeze(0)
>>> acc = Accuracy(
...     torch.log(probs),
...     torch.tensor([1, 1, 0]).unsqueeze(0),
...     torch.tensor([2 / 3]),
... )
>>> print(acc)
(1.0, 2.0)

class speechbrain.utils.Accuracy.AccuracyStats

Bases: object

Module for calculate the overall one-step-forward prediction accuracy.

Example

>>> probs = torch.tensor([[0.9, 0.1], [0.1, 0.9], [0.8, 0.2]]).unsqueeze(0)
>>> stats = AccuracyStats()
>>> stats.append(
...     torch.log(probs),
...     torch.tensor([1, 1, 0]).unsqueeze(0),
...     torch.tensor([2 / 3]),
... )
>>> acc = stats.summarize()
>>> print(acc)
0.5

append(log_probabilities, targets, length=None)

This function is for updating the stats according to the prediction and target in the current batch.

Parameters:

• log_probabilities (torch.Tensor) – Predicted log probabilities (batch_size, time, feature).

• targets (torch.Tensor) – Target (batch_size, time).

• length (torch.Tensor) – Length of target (batch_size,).

summarize()

Computes the accuracy metric.