Line data    Source code

       1             : /*************************************************************************
       2             :  *
       3             :  *  Project
       4             :  *                         _____ _____  __  __ _____
       5             :  *                        / ____|  __ \|  \/  |  __ \
       6             :  *  ___  _ __   ___ _ __ | |  __| |__) | \  / | |__) |
       7             :  * / _ \| '_ \ / _ \ '_ \| | |_ |  ___/| |\/| |  ___/
       8             :  *| (_) | |_) |  __/ | | | |__| | |    | |  | | |
       9             :  * \___/| .__/ \___|_| |_|\_____|_|    |_|  |_|_|
      10             :  *      | |
      11             :  *      |_|
      12             :  *
      13             :  * Copyright (C) Akiel Aries, <akiel@akiel.org>, et al.
      14             :  *
      15             :  * This software is licensed as described in the file LICENSE, which
      16             :  * you should have received as part of this distribution. The terms
      17             :  * among other details are referenced in the official documentation
      18             :  * seen here : https://akielaries.github.io/openGPMP/ along with
      19             :  * important files seen in this project.
      20             :  *
      21             :  * You may opt to use, copy, modify, merge, publish, distribute
      22             :  * and/or sell copies of the Software, and permit persons to whom
      23             :  * the Software is furnished to do so, under the terms of the
      24             :  * LICENSE file. As this is an Open Source effort, all implementations
      25             :  * must be of the same methodology.
      26             :  *
      27             :  *
      28             :  *
      29             :  * This software is distributed on an AS IS basis, WITHOUT
      30             :  * WARRANTY OF ANY KIND, either express or implied.
      31             :  *
      32             :  ************************************************************************/
      33             : #include <openGPMP/ml/trainers.hpp>
      34             : #include <random>
      35             : 
      36             : std::vector<double>
      37           0 : gpmp::ml::Trainers::gradientdesc(const std::vector<std::vector<double>> &X,
      38             :                                  const std::vector<double> &y,
      39             :                                  double alpha,
      40             :                                  int num_iters) {
      41           0 :     int m = X.size();                  // Number of training examples
      42           0 :     int n = X[0].size();               // Number of features
      43           0 :     std::vector<double> theta(n, 0.0); // Initialize parameters
      44             : 
      45             :     // Perform gradient descent
      46           0 :     for (int iter = 0; iter < num_iters; iter++) {
      47             :         std::vector<double> delta_theta(n,
      48           0 :                                         0.0); // Initialize change in parameters
      49             : 
      50             :         // Compute predictions and errors
      51           0 :         for (int i = 0; i < m; i++) {
      52           0 :             double prediction = 0;
      53           0 :             for (int j = 0; j < n; j++) {
      54           0 :                 prediction += theta[j] * X[i][j];
      55             :             }
      56           0 :             double error = prediction - y[i];
      57             : 
      58             :             // Update delta_theta
      59           0 :             for (int j = 0; j < n; j++) {
      60           0 :                 delta_theta[j] += error * X[i][j];
      61             :             }
      62             :         }
      63             : 
      64             :         // Update parameters
      65           0 :         for (int j = 0; j < n; j++) {
      66           0 :             theta[j] -= (alpha / m) * delta_theta[j];
      67             :         }
      68           0 :     }
      69             : 
      70           0 :     return theta;
      71           0 : }
      72             : 
      73           0 : std::vector<double> gpmp::ml::Trainers::stoch_gradientdesc(
      74             :     const std::vector<std::vector<double>> &X,
      75             :     const std::vector<double> &y,
      76             :     double alpha,
      77             :     int num_iters) {
      78           0 :     int m = X.size();                  // Number of training examples
      79           0 :     int n = X[0].size();               // Number of features
      80           0 :     std::vector<double> theta(n, 0.0); // Initialize parameters
      81             : 
      82           0 :     std::default_random_engine generator;
      83           0 :     std::uniform_int_distribution<int> distribution(0, m - 1);
      84             : 
      85           0 :     for (int iter = 0; iter < num_iters; iter++) {
      86             :         int random_index =
      87           0 :             distribution(generator); // Choose a random training example
      88           0 :         double prediction = 0;
      89             : 
      90           0 :         for (int j = 0; j < n; j++) {
      91           0 :             prediction += theta[j] * X[random_index][j];
      92             :         }
      93             : 
      94           0 :         double error = prediction - y[random_index];
      95             : 
      96           0 :         for (int j = 0; j < n; j++) {
      97           0 :             theta[j] -= (alpha * error * X[random_index][j]);
      98             :         }
      99             :     }
     100             : 
     101           0 :     return theta;
     102           0 : }
     103           0 : std::vector<double> gpmp::ml::Trainers::minibatch_gradientdesc(
     104             :     const std::vector<std::vector<double>> &X,
     105             :     const std::vector<double> &y,
     106             :     double alpha,
     107             :     int num_iters,
     108             :     int batch_size) {
     109           0 :     int m = X.size();                  // Number of training examples
     110           0 :     int n = X[0].size();               // Number of features
     111           0 :     std::vector<double> theta(n, 0.0); // Initialize parameters
     112             : 
     113           0 :     std::default_random_engine generator;
     114           0 :     std::uniform_int_distribution<int> distribution(0, m - 1);
     115             : 
     116           0 :     for (int iter = 0; iter < num_iters; iter++) {
     117             :         // Select random batch indices
     118           0 :         std::vector<int> batch_indices;
     119           0 :         for (int i = 0; i < batch_size; i++) {
     120           0 :             batch_indices.push_back(distribution(generator));
     121             :         }
     122             : 
     123             :         // Compute gradient for the batch
     124           0 :         std::vector<double> delta_theta(n, 0.0);
     125           0 :         for (int idx : batch_indices) {
     126           0 :             double prediction = 0;
     127           0 :             for (int j = 0; j < n; j++) {
     128           0 :                 prediction += theta[j] * X[idx][j];
     129             :             }
     130           0 :             double error = prediction - y[idx];
     131             : 
     132           0 :             for (int j = 0; j < n; j++) {
     133           0 :                 delta_theta[j] += (error * X[idx][j]);
     134             :             }
     135             :         }
     136             : 
     137             :         // Update parameters using the gradient of the batch
     138           0 :         for (int j = 0; j < n; j++) {
     139           0 :             theta[j] -= (alpha / batch_size) * delta_theta[j];
     140             :         }
     141           0 :     }
     142             : 
     143           0 :     return theta;
     144           0 : }