pass_generator.h

#pragma once
 
#include <random>
 
#include "proto/parameters.pb.h"
#include "software/ai/passing/cost_function.h"
#include "software/ai/passing/pass_with_rating.h"
#include "software/optimization/gradient_descent_optimizer.hpp"
#include "software/world/world.h"
 
class PassGenerator
{
   public:
    explicit PassGenerator(const TbotsProto::PassingConfig& passing_config);
 
    PassWithRating getBestPass(const World& world,
                               const std::vector<RobotId>& robots_to_ignore = {});
 
   private:
    std::map<RobotId, std::vector<Point>> sampleReceivingPositionsPerRobot(
        const World& world, const std::vector<RobotId>& robots_to_ignore);
 
    PassWithRating optimizeReceivingPositions(
        const World& world,
        const std::map<RobotId, std::vector<Point>>& receiving_positions_map);
 
    // Weights used to normalize the parameters that we pass to GradientDescent
    // (see the GradientDescent documentation for details)
    // These weights are *very* roughly the step that gradient descent will take
    // in each respective dimension for a single iteration. They are tuned to
    // ensure passes converge as fast as possible, but are also as stable as
    // possible
    static constexpr double PASS_SPACE_WEIGHT                          = 0.1;
    std::array<double, NUM_PARAMS_TO_OPTIMIZE> optimizer_param_weights = {
        PASS_SPACE_WEIGHT, PASS_SPACE_WEIGHT};
 
    // The optimizer we're using to find passes
    GradientDescentOptimizer<NUM_PARAMS_TO_OPTIMIZE> optimizer_;
 
    std::map<RobotId, Point> previous_best_receiving_positions_;
 
    // the random seed used to initialize the random number generator
    static constexpr int RNG_SEED = 1010;
 
    // A random number generator for use across the class
    std::mt19937 random_num_gen_;
 
    // Passing configuration
    TbotsProto::PassingConfig passing_config_;
};