/**
 * @file eqpalg/regression/krr.h
 * @brief krr
 * @author Cat (null.null.null@qq.com)
 * @version 0.1
 * @date 2021-09-17
 *
 * Copyright: Baosight Co. Ltd.
 * DO NOT COPY/USE WITHOUT PERMISSION
 *
 */
#pragma once
#include <eqpalg/regression/train_result.h>
#include "mix_cc/algorithm/split.h"
#include <eqpalg/define/sample.h>
#include <vector>
#include <memory>
namespace regression {
using namespace dlib;
/**
 * @brief krr
 * @tparam rows
 * @tparam cols
 */
class Krr {
 public:
  //
  using SampleType = matrix<double>;
  using KernelType = radial_basis_kernel<SampleType>;

 private:
  double split_precent_;
  const size_t dims_;
  std::unique_ptr<decision_function<KernelType>> predict_func_;

  double err_rate_;

 public:
  Krr(size_t dims, const double gama, const std::vector<SampleType>& tr_samples,
      const std::vector<double>& tr_labels)
      : dims_(dims) {
    this->split_precent_ = 0.7;
    krr_trainer<KernelType> trainer;
    trainer.set_kernel(KernelType(gama));
    size_t split_size = tr_samples.size() * (split_precent_);
    auto splitted_sample = mix_cc::split(tr_samples, {split_size});
    auto splitted_label = mix_cc::split(tr_labels, {split_size});
    predict_func_ = std::make_unique<decision_function<KernelType>>(
        trainer.train(splitted_sample[0], splitted_label[0]));
    std::vector<double> loo_values;
    trainer.train(splitted_sample[1], splitted_label[1], loo_values);
    err_rate_ = mean_squared_error(splitted_label[1], loo_values);
  }

  double predict(SamplePoint input) {
    dlib::matrix<double> input_convert(dims_,1);
    for (size_t i = 0; i < dims_; i++) {
      input_convert(i) = input[i];
    }
    return (*predict_func_)(input_convert);
  }
};
}  // namespace regression