34_train_regressor_ellipsoid.cpp

This is an example of how to use the netdem library.

#include "general_net.hpp"
#include "mlpack_utils.hpp"
#include "regression_net.hpp"
#include "utils_math.hpp"
 
using namespace netdem;
using namespace std;
 
void TrainRegressorEllipsoid() {
  string root_dir = "local/examples/netdem/ann_models/ellipsoid_ellipsoid/";
 
  // notes: rows represent features, columns represent data points.
  arma::mat dataset;
  bool loadedDataset = mlpack::data::Load(
      root_dir + "archived/dataset_resolution.txt", dataset, true);
  if (!loadedDataset) {
    cout << "loading data erros ..." << endl;
    return;
  }
  for (int i = 0; i < dataset.n_cols; i++) {
    auto tmp_sgn = Math::Sign(dataset(3, i));
    dataset(3, i) = tmp_sgn * dataset(3, i);
    dataset(4, i) = tmp_sgn * dataset(4, i);
    dataset(5, i) = tmp_sgn * dataset(5, i);
    dataset(6, i) = tmp_sgn * dataset(6, i);
    dataset(7, i) = dataset(7, i) * 40.0;
  }
  cout << "finished loading data ..." << endl;
 
  // split traning and testing data, prepare inputs and outputs
  arma::mat train_data, test_data;
  mlpack::data::Split(dataset, train_data, test_data, 0.1);
  int num_trains = train_data.n_cols;
  int num_tests = test_data.n_cols;
 
  int num_x{7}, num_y{7};
  arma::mat train_x = train_data.rows(0, num_x - 1);
  arma::mat test_x = test_data.rows(0, num_x - 1);
 
  arma::mat train_y = train_data.rows(num_x, num_x + num_y - 1);
  arma::mat test_y = test_data.rows(num_x, num_x + num_y - 1);
  cout << "finished preparing training and testing data ..." << endl;
 
  // create the ann model
  RegressionNet model;
  int num_nodes = 50, num_hidden_layers = 5;
  model.AddLayer(MLPackUtils::LayerType::Linear, num_nodes);
  model.AddLayer(MLPackUtils::LayerType::LeakyReLU);
  for (int i = 0; i < num_hidden_layers - 1; i++) {
    model.AddLayer(MLPackUtils::LayerType::Linear, num_nodes);
    model.AddLayer(MLPackUtils::LayerType::LeakyReLU);
  }
  model.AddLayer(MLPackUtils::LayerType::Linear, train_y.n_rows);
  cout << "finished creating ann model ..." << endl;
 
  // set training parameters
  model.epochs = 100;
  model.step_size = 0.001;
  model.batch_size = 128;
  model.stop_tol = 1.0e-10;
  model.decay_rate_moment = 0.99;
  model.decay_rate_norm = 0.99;
 
  double test_accuracy_best{1.0e8};
  for (int i = 0; i < 10; i++) {
    model.Train(train_x, train_y);
    cout << endl << "finished training ..." << i << endl;
 
    // check the training and testing results
    auto pred_y = model.Predict(train_x);
    double train_accuracy = MLPackUtils::GetMAE(train_y, pred_y);
    cout << "train mae: " << train_accuracy << endl;
 
    pred_y = model.Predict(test_x);
    double test_accuracy = MLPackUtils::GetMAE(pred_y, test_y);
    cout << "test mae: " << test_accuracy << endl << endl;
 
    // save the model
    model.Save(root_dir + "training/ann_regressor" + to_string(i) + ".xml",
               "resolution");
 
    if (test_accuracy < test_accuracy_best) {
      test_accuracy_best = test_accuracy;
      model.Save(root_dir + "ann_regressor.xml", "resolution");
    }
  }
}