Regression Metrics

This section documents the regression metrics available in pytorch_tabnet.metrics.

Mean Absolute Error (MAE)

Measures the average magnitude of the errors in a set of predictions, without considering their direction.

\[\mathrm{MAE} = \frac{1}{N} \sum_{i=1}^N |y_i - \hat{y}_i|\]

Example:

import torch
from pytorch_tabnet.metrics import MAE
metric = MAE()
y_true = torch.tensor([3, -0.5, 2, 7])
y_pred = torch.tensor([2.5, 0.0, 2, 8])
print(metric(y_true, y_pred))  # Output: 0.5

Mean Squared Error (MSE)

Measures the average of the squares of the errors between actual and predicted values.

\[\mathrm{MSE} = \frac{1}{N} \sum_{i=1}^N (y_i - \hat{y}_i)^2\]

Example:

import torch
from pytorch_tabnet.metrics import MSE
metric = MSE()
y_true = torch.tensor([3, -0.5, 2, 7])
y_pred = torch.tensor([2.5, 0.0, 2, 8])
print(metric(y_true, y_pred))  # Output: 0.375

Root Mean Squared Error (RMSE)

The square root of the mean squared error, providing error in the same units as the target variable.

\[\mathrm{RMSE} = \sqrt{\frac{1}{N} \sum_{i=1}^N (y_i - \hat{y}_i)^2}\]

Example:

import torch
from pytorch_tabnet.metrics import RMSE
metric = RMSE()
y_true = torch.tensor([3, -0.5, 2, 7])
y_pred = torch.tensor([2.5, 0.0, 2, 8])
print(metric(y_true, y_pred))  # Output: 0.612...

Root Mean Squared Logarithmic Error (RMSLE)

Measures the ratio between the true and predicted values, less sensitive to large errors when both values are large.

\[\mathrm{RMSLE} = \sqrt{\frac{1}{N} \sum_{i=1}^N \left( \log(y_i + 1) - \log(\hat{y}_i + 1) \right)^2}\]

Example:

import torch
from pytorch_tabnet.metrics import RMSLE
metric = RMSLE()
y_true = torch.tensor([3, 5, 2.5, 7])
y_pred = torch.tensor([2.5, 5, 4, 8])
print(metric(y_true, y_pred))  # Output: 0.120...