Visualization of Loss Functions

Loss Functions

library(vistool)

The VisualizerLossFunction class visualizes loss functions. Loss functions are wrapped in LossFunction objects. The package includes a few predefined loss functions that are used for classification and regression.

as.data.table(dict_loss)
#> Key: <key>
#>              key             label properties
#>           <char>            <char>     <list>
#> 1: cross-entropy     Cross-Entropy    classif
#> 2:         hinge        Hinge Loss    classif
#> 3:         huber        Huber Loss       regr
#> 4:         l1_ae L1 Absolute Error       regr
#> 5:         l2_se  L2 Squared Error       regr
#> 6:      log-cosh     Log-Cosh Loss       regr

To get a loss function from the dictionary, use the lss() function We retrieve the hinge loss function for classification.

loss_function = lss("hinge")

Visualization

To visualize the loss function, use the as_visualizer() function. The y_pred arguments contains $y * f(x)$ and y_true the true value $y$.

vis = as_visualizer(loss_function, y_pred = seq(-4, 4), y_true = 1)

Use the plot() method to plot the loss function.

vis$plot()

Regression Loss Functions

Classification Loss Functions