#include <eqpalg/oneClassSvm/frame.hpp>
namespace oneClassSvm {

/**
 * @brief 训练SVM 将结果赋值给svm_result_ 并返回结果的引用
 * @tparam dims
 * @param  tr_samples       My Param doc
 * @return TrainResult
 */
template <size_t dims>
TrainResult& Frame<dims>::train_get_df_one_class_svm(SamplesType& tr_samples) {
  svm_result_ = decision_function_get(tr_samples);
  return this->svm_result_;
}
/**
 * @brief 预测
 * @param  data             My Param doc
 * @return double  越小越异常
 */
template <size_t dims>
double Frame<dims>::predict(ExchangedMetaData& data) {
  matrix<double, 0, 1> tmp;
  for (size_t i = 0; i < data.size(); i++) {
    tmp(i) = data[i];
  }
  return svm_result_(tmp);
}
/**
 * @brief 载入数据
 * @tparam dims
 * @param  data             My Param doc
 * @return int
 */
template <size_t dims>
int Frame<dims>::load(const InData& data) {
  SamplesType tr_samples;
  // 如果不存在数据，退出
  if (data.empty()) {
    return -1;
  }
  // 提前保留样本
  tr_samples.reserve(data.size());
  // 把输入转成d—lib样本
  for (const auto& meta_data : data) {
    matrix<double, 0, 1> tmp(dims);
    for (size_t i = 0; i < meta_data.size(); i++) {
      tmp(i) = meta_data[i];
    }
    tr_samples.emplace_back(tmp);
  }
  auto df = train_get_df_one_class_svm(tr_samples);
  double th = 0;
  for (auto x : tr_samples) th += df(x);
  th = th / tr_samples.size();
  standard_error = -fabs(th * 100);
  // standard_error =
  return 0;
}
template <size_t dims>
bool Frame<dims>::valid() {
  if (standard_error < 0) {
    return true;
  }
  return false;
}

 TrainResult decision_function_get(SamplesType samples) 
    {
        svm_one_class_trainer<kernel_type> trainer;
        trainer.set_kernel(kernel_type(4.0));
        return trainer.train(samples);
    }
}  // namespace oneClassSvm

template class oneClassSvm::Frame<1>;
template class oneClassSvm::Frame<2>;
template class oneClassSvm::Frame<3>;