C++ Simplex::neldermead方法代码示例

double Handler::optimise_single(vector<double> &parameters, vector<vector<double> > &results, vector<vector<vector<double> > > &allResults, int& itr){
  double SSE = 99999999999999.9;
  double tempSSE;
  vector<double> tempPar, seedParams;
  Simplex simplex;

  parameters.clear();
  allResults.clear();
  parameters = generate_seed_parameters();
  for(int index = 0;index<20;index++){
    seedParams.clear();
    seedParams = generate_seed_parameters();
    //printcon(seedParams);
    if(this->useMLE == false) tempPar = simplex.neldermead(&Handler::fitEpidemicsMLE, *this,  seedParams, itr);
    else tempPar = simplex.neldermead(&Handler::fitEpidemics, *this,  seedParams, itr);
    // Store the SSE value for this
    tempSSE = fitEpidemics(tempPar);
    if(tempSSE < SSE){
      SSE = tempSSE;
      parameters=tempPar;
    }
    cout << "." << flush;
  }
  results = ode_solve(parameters);
  allResults = ode_solve_separate(parameters);
  //allResults.clear();
  cout << endl;
  return SSE;
}

/* Optimisation procedure. Takes a reference to parameters, combined results, component results and number of iterations taken. The function updates these
   arguments with the results of the Nelder-Mead optimisation procedure. */
double Handler::optimiseEpidemics(vector<double> &parameters, vector<vector<double> > &results, vector<vector<vector<double> > > &allResults, int& itr){
  double SSE = 99999999999999999.9;
  double tempSSE;
  vector<double> tempPar, seedParams;
  Simplex simplex;
  int iterations = 10001;

  // If no epidemics have yet been detected, use the mean of the current data as the 
  // current model and return the corresponding SSE.
  if(epidemics.size() == 0){
    //baseModel = base_model(temp_data);
    results = base_model(temp_data);
    SSE = calculate_SSE(results, temp_data);
    parameters.clear();
    allResults.clear();
    allResults.push_back(baseModel);
    return(SSE);
  }
  srand(time(NULL));
  parameters = generate_seed_parameters();

  // If there are epidemics to be fitted, perform X random fits and keep best fitting model
  for(int index=0;index<10;index++){
    // Create random seed parameters
    seedParams.clear();      
    seedParams = generate_seed_parameters();
    
    while(fitEpidemics(seedParams) != fitEpidemics(seedParams)){
      seedParams.clear();            
      seedParams = generate_seed_parameters();
    }
    
    // Get the optimised parameters from nelder mead algorithm
    tempPar = simplex.neldermead(&Handler::fitEpidemics, *this,  seedParams, iterations);
    // Store the SSE value for this
 
    tempSSE = fitEpidemics(tempPar);
    //if(tempSSE == tempSSE) cout << "Temp RSquare: " << (1-tempSSE/(SStot(temp_data,1))) << endl;
 
    // If this SSE value is better than the previous, store it and the corresponding parameters
    if(tempSSE < SSE){
      itr = iterations;
      SSE = tempSSE;
      parameters=tempPar;
    }
    cout << "." << flush;
  }

  // Get the combined values from these parameters, as well as a vector of 
  // each sub-epidemic
  results = ode_solve(parameters);
   
  allResults.clear();
  allResults = ode_solve_separate(parameters);
  cout << endl;
  return(SSE);

}

double Handler::optimiseEpidemics(vector<double> &parameters, vector<vector<double> > &results, vector<vector<vector<double> > > &allResults, int& itr){
  double SSE = 99999999999.9;
  double tempSSE;
  vector<double> tempPar, seedParams;
  vector<vector<vector<double> > > tempAll;
  Simplex simplex;
  int iterations = 10001;
  // If no epidemics have yet been detected, use the mean of the current data as the 
  // current model and return the corresponding SSE.
  if(epidemics.size() == 0){
    results = this->baseModel;
    SSE = calculate_SSE(results, temp_data);
    parameters.clear();
    allResults.clear();
    allResults.push_back(results);
    return(SSE);
  }
  parameters = generate_seed_parameters();
  // If there are epidemics to be fitted, perform 40 random fits and keep best fitting model
  for(int index=0;index<40;index++){    
            
    // Clear the temporary seed parameters and seed rand
   

    // Create a list of random seed parameters
    seedParams.clear();      
    seedParams = generate_seed_parameters();
    
    // Get the optimised parameters from nelder mead algorithm
    if(this->useMLE == false) tempPar = simplex.neldermead(&Handler::fitEpidemicsMLE, *this,  seedParams, iterations);
    else tempPar = simplex.neldermead(&Handler::fitEpidemics, *this,  seedParams, iterations);
    // Store the SSE value for this
    tempSSE = fitEpidemics(tempPar);
	
    // If this SSE value is better than the previous, store it and the
    // corresponding parameters
    if(tempSSE < SSE){
      itr = iterations;
      SSE = tempSSE;
      parameters=tempPar;
    }
    cout << "." << flush;
  }
 
  // Get the combined values from these parameters, as well as a vector of 
  // each sub-epidemic
  results = ode_solve(parameters);
  
  tempAll = ode_solve_separate(parameters);
  
  allResults.clear();
  allResults.push_back(this->baseModel);
  
  for(unsigned int x = 0;x<tempAll.size();++x){
    allResults.push_back(tempAll[x]);
  }
  
  cout << endl;
  
  return(SSE);

}