--- title: Losses utils keywords: fastai sidebar: home_sidebar summary: "Generic loss class." description: "Generic loss class." nb_path: "nbs/losses__utils.ipynb" ---
{% raw %}
{% endraw %} {% raw %}
{% endraw %} {% raw %}

class LossFunction[source]

LossFunction(loss_name:str, seasonality:Optional[int]=None, percentile:Union[List[int], int, NoneType]=None, level_variability_penalty:Optional[int]=None)

{% endraw %} {% raw %}
{% endraw %} {% raw %}

LossFunction.__call__[source]

LossFunction.__call__(y:Tensor, y_hat:Tensor, mask:Optional[Tensor]=None, y_insample:Optional[Tensor]=None, levels:Optional[Tensor]=None)

Returns loss according to loss_name.

{% endraw %} {% raw %}
{% endraw %}

Usage

Multi-Quantile losses

{% raw %}
loss = LossFunction('MQ', percentile=[30, 60])
y = t.normal(0, 1, size=(10, 10))
y_hat = t.normal(0, 1, size=(10, 10, 2))

loss(y, y_hat)
tensor(0.5067)
{% endraw %}

Mean losses

{% raw %}
loss = LossFunction('MAE')
y = t.normal(0, 1, size=(10, 10))
y_hat = t.normal(0, 1, size=(10, 10))

loss(y, y_hat)
tensor(1.1782)
{% endraw %}