Tweedie Deviance Score¶

Module Interface¶

class torchmetrics.TweedieDevianceScore(power=0.0, **kwargs)[source]

Computes the Tweedie Deviance Score between targets and predictions:

$deviance\_score(\hat{y},y) = \begin{cases} (\hat{y} - y)^2, & \text{for }p=0\\ 2 * (y * log(\frac{y}{\hat{y}}) + \hat{y} - y), & \text{for }p=1\\ 2 * (log(\frac{\hat{y}}{y}) + \frac{y}{\hat{y}} - 1), & \text{for }p=2\\ 2 * (\frac{(max(y,0))^{2 - p}}{(1 - p)(2 - p)} - \frac{y(\hat{y})^{1 - p}}{1 - p} + \frac{( \hat{y})^{2 - p}}{2 - p}), & \text{otherwise} \end{cases}$

where $y$ is a tensor of targets values, $\hat{y}$ is a tensor of predictions, and $p$ is the power.

As input to forward and update the metric accepts the following input:

preds (Tensor): Predicted float tensor with shape (N,...)
target (Tensor): Ground truth float tensor with shape (N,...)

As output of forward and compute the metric returns the following output:

deviance_score (Tensor): A tensor with the deviance score

Parameters

power¶ (float) –
- power < 0 : Extreme stable distribution. (Requires: preds > 0.)
- power = 0 : Normal distribution. (Requires: targets and preds can be any real numbers.)
- power = 1 : Poisson distribution. (Requires: targets >= 0 and y_pred > 0.)
- 1 < p < 2 : Compound Poisson distribution. (Requires: targets >= 0 and preds > 0.)
- power = 2 : Gamma distribution. (Requires: targets > 0 and preds > 0.)
- power = 3 : Inverse Gaussian distribution. (Requires: targets > 0 and preds > 0.)
- otherwise : Positive stable distribution. (Requires: targets > 0 and preds > 0.)
kwargs¶ (Any) – Additional keyword arguments, see Advanced metric settings for more info.

Example

>>> from torchmetrics import TweedieDevianceScore
>>> targets = torch.tensor([1.0, 2.0, 3.0, 4.0])
>>> preds = torch.tensor([4.0, 3.0, 2.0, 1.0])
>>> deviance_score = TweedieDevianceScore(power=2)
>>> deviance_score(preds, targets)
tensor(1.2083)

Initializes internal Module state, shared by both nn.Module and ScriptModule.

Functional Interface¶

torchmetrics.functional.tweedie_deviance_score(preds, targets, power=0.0)[source]

Computes the Tweedie Deviance Score between targets and predictions:

$deviance\_score(\hat{y},y) = \begin{cases} (\hat{y} - y)^2, & \text{for }p=0\\ 2 * (y * log(\frac{y}{\hat{y}}) + \hat{y} - y), & \text{for }p=1\\ 2 * (log(\frac{\hat{y}}{y}) + \frac{y}{\hat{y}} - 1), & \text{for }p=2\\ 2 * (\frac{(max(y,0))^{2 - p}}{(1 - p)(2 - p)} - \frac{y(\hat{y})^{1 - p}}{1 - p} + \frac{( \hat{y})^{2 - p}}{2 - p}), & \text{otherwise} \end{cases}$

where $y$ is a tensor of targets values, $\hat{y}$ is a tensor of predictions, and $p$ is the power.

Parameters

preds¶ (Tensor) – Predicted tensor with shape (N,...)
targets¶ (Tensor) – Ground truth tensor with shape (N,...)
power¶ (float) –
- power < 0 : Extreme stable distribution. (Requires: preds > 0.)
- power = 0 : Normal distribution. (Requires: targets and preds can be any real numbers.)
- power = 1 : Poisson distribution. (Requires: targets >= 0 and y_pred > 0.)
- 1 < p < 2 : Compound Poisson distribution. (Requires: targets >= 0 and preds > 0.)
- power = 2 : Gamma distribution. (Requires: targets > 0 and preds > 0.)
- power = 3 : Inverse Gaussian distribution. (Requires: targets > 0 and preds > 0.)
- otherwise : Positive stable distribution. (Requires: targets > 0 and preds > 0.)

Example

>>> from torchmetrics.functional import tweedie_deviance_score
>>> targets = torch.tensor([1.0, 2.0, 3.0, 4.0])
>>> preds = torch.tensor([4.0, 3.0, 2.0, 1.0])
>>> tweedie_deviance_score(preds, targets, power=2)
tensor(1.2083)

Return type: Tensor