Mean absolute error

The average loss per example when L1 loss is used. Calculate MAE as follows:¹

1. Calculate the L1 loss for a batch.
2. Divide the L1 loss by the number of examples in the batch.

$$\text{MAE} = \frac{1}{n}\sum_{i=0}^{n} | y_i - \hat{y}_i |$$

where:

• $n$ is the number of examples.
• $y$ is the actual value of the label.
• $\hat{y}$ is the value that the model predicts for $y$.

See also

• Mean squared error
• Root mean squared error

Footnotes

1. developers.google.com/machine-learning/glossary#MAE ↩