Nous.Eval.Optimizer.Strategy behaviour (nous v0.15.8)

View Source

Behaviour for optimization strategies.

Strategies define how to explore the search space to find optimal configurations.

Implementing a Custom Strategy

defmodule MyStrategy do
  @behaviour Nous.Eval.Optimizer.Strategy

  @impl true
  def run(suite, search_space, metric, maximize, opts) do
    # Your optimization logic here
    # Return {:ok, [trial, ...]} or {:error, reason}
  end
end

Built-in Strategies

Summary

Types

trial()

Callbacks

run(suite, search_space, metric, maximize, opts)

Run the optimization strategy.

Types

trial()

@type trial() :: Nous.Eval.Optimizer.trial()

Callbacks

run(suite, search_space, metric, maximize, opts)

@callback run(
  suite :: Nous.Eval.Suite.t(),
  search_space :: Nous.Eval.Optimizer.SearchSpace.t(),
  metric :: Nous.Eval.Optimizer.metric(),
  maximize :: boolean(),
  opts :: keyword()
) :: {:ok, [trial()]} | {:error, term()}

Run the optimization strategy.

Parameters

  • suite - The evaluation suite to optimize
  • search_space - The parameter search space
  • metric - The metric to optimize
  • maximize - Whether to maximize (true) or minimize (false)
  • opts - Strategy-specific options

Returns

  • {:ok, trials} - List of all trials run
  • {:error, reason} - If optimization fails