Skip to contents

Activation Functions Specification Helper

Source: R/act-fun.R
act_funs.Rd

This function is a DSL function, kind of like ggplot2::aes(), that helps to specify activation functions for neural network layers. It validates that activation functions exist in torch and that any parameters match the function's formal arguments.

Usage

act_funs(...)

Arguments

...

Activation function specifications. Can be:

  • Bare symbols: relu, tanh

  • Character strings (simple): "relu", "tanh"

  • Character strings (with params): "softshrink(lambda = 0.1)", "rrelu(lower = 1/5, upper = 1/4)"

  • Named with parameters: softmax = args(dim = 2L)

  • Indexed syntax (named): softshrink[lambd = 0.2], rrelu[lower = 1/5, upper = 1/4]

  • Indexed syntax (unnamed): softshrink[0.5], elu[0.5]

Value

A vctrs vector with class "activation_spec" containing validated activation specifications.

Examples

act_funs(relu, sigmoid)
#> <activation_spec[2]>
#>                 
#>    relu sigmoid 
act_funs(relu, softshrink[lambd = 0.5], elu)
#> <activation_spec[3]>
#>                
#> relu  0.5  elu 
act_funs(softmax = args(dim = 2L))
#> Warning: `args()` was deprecated in kindling 0.3.0.
#> ℹ Use indexed syntax for parametric activation functions, e.g. `<softplus[beta
#>   = 0.5]>`.
#> <activation_spec[1]>
#> softmax 
#>       2