sample.cc

// Copyright 2017 The Fuchsia Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.

#include <memory>
#include <random>
#include <iostream>

#include "lossmin/eigen-types.h"
#include "lossmin/losses/loss-function.h"
#include "lossmin/losses/inner-product-loss-function.h"
#include "minimizers/gradient-evaluator.h"
#include "minimizers/loss-minimizer.h"
#include "minimizers/parallel-boosting-with-momentum.h"

using namespace lossmin;

int main(int argc, char **argv) {
  LossFunction *loss_func = new LinearRegressionLossFunction();

  const int rows_num = 10000;
  const int features_num = 3;


  InstanceSet instances(rows_num, features_num);

  std::default_random_engine generator;
  std::uniform_real_distribution<float> distribution(-1.0,1.0);

  for (int row = 0; row < rows_num; row++) {
    for (int feature = 0; feature < features_num; feature++) {
      instances.insert(row, feature) = distribution(generator);
    }
  }

  Weights real_weights(features_num, 1);
  for (int feature = 0; feature < features_num; feature++) {
    real_weights.coeffRef(feature, 0) = distribution(generator);
  }
  std::cout << "Real Weights:" << std::endl << real_weights << std::endl;

  LabelSet labels = instances * real_weights;

  GradientEvaluator gradient_evaluator(instances, labels, loss_func);
  LossMinimizer *minimizer = new ParallelBoostingWithMomentum(
      0.0, 0.0, gradient_evaluator);
  minimizer->Setup();

  Weights weights(features_num, 1);
  std::vector<float> loss;
  minimizer->Run(5000, &weights, &loss);
  std::cout << "Learned Weights:" << std::endl << weights << std::endl;
}