Evolvable protocol implementation for hyperparameter maps.

Supports schema-driven evolution of mixed-type parameters:

• Floats with linear or log-scale bounds
• Integers with min/max bounds
• Enums (categorical choices)
• Lists of values with length constraints

Schema Format

Bounds can be specified as either ranges (min..max) or tuples ({min, max}). Note: Ranges only work for integer bounds (Elixir requirement). Use tuples for float bounds.

%{
  learning_rate: {:float, {1.0e-5, 1.0e-1}, :log},
  hidden_layers: {:list, {:int, 16..256}, length: {1, 3}},
  dropout_rate: {:float, {0.0, 0.6}, :linear},
  activation: {:enum, [:relu, :tanh, :gelu]},
  batch_size: {:enum, [16, 32, 64, 128]}
}

Usage

schema = %{learning_rate: {:float, {0.001, 0.1}, :log}}
hparams = Jido.Evolve.Evolvable.HParams.new(schema)
# => %{learning_rate: 0.0032}