Line data    Source code

       1             : //* This file is part of the MOOSE framework
       2             : //* https://mooseframework.inl.gov
       3             : //*
       4             : //* All rights reserved, see COPYRIGHT for full restrictions
       5             : //* https://github.com/idaholab/moose/blob/master/COPYRIGHT
       6             : //*
       7             : //* Licensed under LGPL 2.1, please see LICENSE for details
       8             : //* https://www.gnu.org/licenses/lgpl-2.1.html
       9             : #ifdef MOOSE_LIBTORCH_ENABLED
      10             : 
      11             : #pragma once
      12             : 
      13             : #include "ActiveLearningReporterBase.h"
      14             : #include "ActiveLearningGaussianProcess.h"
      15             : #include "GaussianProcessSurrogate.h"
      16             : #include "SurrogateModelInterface.h"
      17             : 
      18             : class ActiveLearningGPDecision : public ActiveLearningReporterTempl<Real>,
      19             :                                  public SurrogateModelInterface
      20             : {
      21             : public:
      22             :   static InputParameters validParams();
      23             :   ActiveLearningGPDecision(const InputParameters & parameters);
      24             : 
      25             :   /// Access the number of training samples
      26           8 :   const int & getTrainingSamples() const { return _n_train; }
      27             : 
      28             : protected:
      29             :   /**
      30             :    * This is where most of the computations happen:
      31             :    *   - Data is accumulated for training
      32             :    *   - GP models are trained
      33             :    *   - Decision is made whether more data is needed for GP training
      34             :    */
      35             :   virtual void preNeedSample() override;
      36             : 
      37             :   /**
      38             :    * Based on the computations in preNeedSample, the decision to get more data is passed and results
      39             :    * from the GP fills @param val
      40             :    *
      41             :    * @param row Input parameters to the model
      42             :    * @param local_ind Current processor row index
      43             :    * @param global_ind All processors row index
      44             :    * @param val Output predicted by either the LF model + GP correction or the HF model
      45             :    * @return bool Whether a full order model evaluation is required
      46             :    */
      47             :   virtual bool needSample(const std::vector<Real> & row,
      48             :                           dof_id_type local_ind,
      49             :                           dof_id_type global_ind,
      50             :                           Real & val) override;
      51             : 
      52             :   /**
      53             :    * Make decisions whether to call the full model or not based on
      54             :    * GP prediction and uncertainty.
      55             :    *
      56             :    * @return bool Whether a full order model evaluation is required
      57             :    */
      58             :   virtual bool facilitateDecision();
      59             : 
      60             :   /**
      61             :    * This sets up data for re-training the GP.
      62             :    *
      63             :    * @param inputs Matrix of inputs for the current step
      64             :    * @param outputs Vector of outputs for the current step
      65             :    */
      66             :   virtual void setupData(const std::vector<std::vector<Real>> & inputs,
      67             :                          const std::vector<Real> & outputs);
      68             : 
      69             :   /**
      70             :    * This method evaluates the active learning acquisition function and returns bool
      71             :    * that indicates whether the GP model failed.
      72             :    *
      73             :    * @param gp_mean Mean of the gaussian process model
      74             :    * @param gp_mean Standard deviation of the gaussian process model
      75             :    * @return bool If the GP model failed
      76             :    */
      77             :   bool learningFunction(const Real & gp_mean, const Real & gp_std) const;
      78             : 
      79             :   /// The learning function for active learning
      80             :   const MooseEnum & _learning_function;
      81             :   /// The learning function threshold
      82             :   const Real & _learning_function_threshold;
      83             :   /// The learning function parameter
      84             :   const Real & _learning_function_parameter;
      85             : 
      86             :   /// Store all the input vectors used for training
      87             :   std::vector<std::vector<Real>> _inputs_batch;
      88             :   /// Store all the outputs used for training
      89             :   std::vector<Real> _outputs_batch;
      90             : 
      91             :   /// The active learning GP trainer that permits re-training
      92             :   const ActiveLearningGaussianProcess & _al_gp;
      93             :   /// The GP evaluator object that permits re-evaluations
      94             :   const SurrogateModel & _gp_eval;
      95             : 
      96             :   /// Flag samples when the GP fails
      97             :   std::vector<bool> & _flag_sample;
      98             : 
      99             :   /// Number of initial training points for GP
     100             :   const int _n_train;
     101             : 
     102             :   /// Storage for the input vectors to be transferred to the output file
     103             :   std::vector<std::vector<Real>> & _inputs;
     104             : 
     105             :   /// Broadcast the GP mean prediciton to JSON
     106             :   std::vector<Real> & _gp_mean;
     107             :   /// Broadcast the GP standard deviation to JSON
     108             :   std::vector<Real> & _gp_std;
     109             : 
     110             :   /// GP pass/fail decision
     111             :   bool _decision;
     112             : 
     113             :   /// Reference to global input data requested from base class
     114             :   const std::vector<std::vector<Real>> & _inputs_global;
     115             :   /// Reference to global output data requested from base class
     116             :   const std::vector<Real> & _outputs_global;
     117             : };
     118             : 
     119             : #endif