Neurons-per-layer heuristic (Geron Ch 10, morie.fn hmnpl)

Constant width w (default 2d) across L hidden layers, with the parameter count per layer. This is the LAYER-ALGEBRA route that the parity test checks against a direct count.

Usage

morie_geron_neurons_per_layer(
  n_features,
  n_layers = 1,
  n_outputs = 1,
  width = NULL
)

Arguments

n_features

Input width.

n_layers

Hidden layers.

n_outputs

Output width.

width

Override for w.

Value

List with width, width_range, n_layers, n_parameters, parameters_per_layer, estimate, n.

Examples

morie_geron_neurons_per_layer(n_features = 5L)
#> $width
#> [1] 10
#> 
#> $width_range
#> [1]  5 10
#> 
#> $n_layers
#> [1] 1
#> 
#> $n_parameters
#> [1] 71
#> 
#> $parameters_per_layer
#> [1] 60 11
#> 
#> $estimate
#> [1] 10
#> 
#> $n
#> [1] 5
#> 
#> $method
#> [1] "Constant-width hidden-layer heuristic with parameter count"
#>