using OECS;

namespace OECS.PlayTest;

/// <summary>
/// Abstract agent that picks the action with the highest score.
/// When <see cref="ScoreAction"/> returns 0 for all actions (the default),
/// the agent behaves as a uniform random agent over legal actions.
/// Ties are broken randomly.
/// </summary>
public abstract class GreedyAgent<TDecision> : IAgent<TDecision>
{
    private static readonly Random _rng = new();

    /// <summary>
    /// Returns the list of currently legal actions.
    /// </summary>
    protected abstract List<TDecision> GetLegalActions(World world);

    /// <summary>
    /// Scores an action. Defaults to 0 (uniform random).
    /// </summary>
    protected virtual float ScoreAction(World world, TDecision action) => 0f;

    public TDecision Decide(World world)
    {
        var actions = GetLegalActions(world);
        if (actions.Count == 0)
            throw new InvalidOperationException(
                $"{GetType().Name}: no legal actions available");

        if (actions.Count == 1)
            return actions[0];

        float bestScore = float.MinValue;
        var bestActions = new List<TDecision>();
        foreach (var action in actions)
        {
            float score = ScoreAction(world, action);
            if (score > bestScore)
            {
                bestScore = score;
                bestActions.Clear();
                bestActions.Add(action);
            }
            else if (score == bestScore)
            {
                bestActions.Add(action);
            }
        }

        return bestActions[_rng.Next(bestActions.Count)];
    }
}