file frontends/classy_2_9_4.cpp

[No description available] More…

Detailed Description

Author:

Date:

  • 2019 June
  • 2020 May,Aug
  • 2019 June
  • 2019 July
  • 2021 January

Frontend source for the classy backend.

Authors (add name and date if you modify):

Source code 

//   GAMBIT: Global and Modular BSM Inference Tool
//   *********************************************
///  \file
///
///  Frontend source for the classy backend.
///
///  *********************************************
///
///  Authors (add name and date if you modify):
///
///  \author Janina Renk
///          (janina.renk@fysik.su.se)
///  \date 2019 June
///  \date 2020 May,Aug
///
///  \author Sanjay Bloor
///          (sanjay.bloor12@imperial.ac.uk)
///  \date 2019 June
///
///  \author Patrick Stoecker
///          (stoecker@physik.rwth-aachen.de)
///  \date 2019 July
///  \date 2021 January
///
///  *********************************************

#include "gambit/Backends/frontend_macros.hpp"
#include "gambit/Backends/frontends/classy_2_9_4.hpp"

#ifdef HAVE_PYBIND11

  #include "gambit/Utils/begin_ignore_warnings_pybind11.hpp"
  #include <pybind11/stl.h>
  #include <pybind11/stl_bind.h>
  #include <pybind11/functional.h>
  #include "gambit/Utils/end_ignore_warnings.hpp"

  BE_NAMESPACE
  {
    // use convenience function to cast a numpy array into a std::vector
    using Backends::cast_np_to_std;

    pybind11::object static cosmo;
    // save input dictionary from CLASS run from
    // previously computed point
    pybind11::dict static prev_input_dict;

    // return cosmo object. Need to pass this to MontePython for Likelihoods calculations
    pybind11::object get_classy_cosmo_object()
    {
      return cosmo;
    }

    // return backend directory of CLASS
    std::string get_classy_backendDir()
    {
      return backendDir;
    }

    /// test if two dictionaries contain exactly the same values for all keys
    /// return true if so, false if there is at least one different value
    bool compare_dicts(pybind11::dict classy_input, pybind11::dict prev_input_dict)
    {
      for (auto it : classy_input)
      {
         // test if any pointer is being passed to CLASS -- if so you'd need to compare the contents
         // of the vectors, so just recompute by default in that case (atm the options that pass pointers
         // can be identified by searching for 'array' or 'pointer_to' in the CLASS input dict. If more of
         // these cases are implemented the checks have to be added here.)
         if(it.first.cast<std::string>().find("array") != std::string::npos){return false;}
         if(it.first.cast<std::string>().find("pointer_to") != std::string::npos){return false;}

         // return false if unequal values are found
         if (classy_input[it.first].cast<std::string>() != prev_input_dict[it.first].cast<std::string>()){return false;}
      }
      // all key values are identical --> no need to run class again, yay!
      return true;
    }

    /// routine to check class input for consistency. If a case is not treated here class will
    /// throw an error saying that one input parameter was not read. Checking for some specific cases
    /// here allows us to give more informative error messages and fix suggestions
    void class_input_consistency_checks(pybind11::dict classy_input)
    {

      std::ostringstream errMssg;
      // one thing that can go wrong is that the primordial power spectrum is requested but
      // no observable depending on perturbations is asked for. In that case, CLASS will only
      // derive background quantities and an error occurs when trying to access quantities
      // depending on perturbations  =>
      // check if "modes" input is set while "output" is not set
      if(classy_input.contains("modes") and not classy_input.contains("output"))
      {
        errMssg << "You are calling CLASS asking for the following modes to be computed : "<< pybind11::repr(classy_input["modes"]);
        errMssg << "\nHowever, you did not request any output that requires computing any perturbation spectra.\nHence CLASS";
        errMssg << " will not read the input 'modes' and won't run. Add the CLASS input parameter 'output' requesting";
        errMssg << " a spectrum to be computed to the yaml file as run option, e.g. \n  - capability: classy_baseline_params\n";
        errMssg << "    options:\n      classy_dict:\n        output: tCl";
        backend_error().raise(LOCAL_INFO,errMssg.str());
      }
    }

    // getter functions to return a bunch of CLASS outputs. This is here in the frontend
    // to make the capabilities inside CosmoBit independent of types that depend on the
    // Boltzmann solver in use

    // get the lensed Cl.
    std::vector<double> class_get_lensed_cl(std::string spectype)
    {
      // Get dictionary containing all (lensed) Cl spectra
      map_str_pyobj cl_dict = cosmo.attr("lensed_cl")().cast<map_str_pyobj>();

      // Get only the relevant Cl as np array and steal the pointer to its data.
      pybind11::object cl_array_obj = cl_dict[spectype];
      pybind11::array_t<double> cl_array = pybind11::cast<pybind11::array_t<double>>(cl_array_obj);

      // Create the vector to return
      std::vector<double> result = cast_np_to_std(cl_array);

      // cl = 0 for l = 0,1
      result.at(0) = 0.;
      result.at(1) = 0.;

      return result;
    }

    // get the raw (unlensed) Cl.
    std::vector<double> class_get_unlensed_cl(std::string spectype)
    {
      // Get dictionary containing the raw (unlensed) Cl spectra
      map_str_pyobj cl_dict = cosmo.attr("raw_cl")().cast<map_str_pyobj>();

      // Get only the relevant Cl as np array and steal the pointer to its data.
      pybind11::object cl_array_obj = cl_dict[spectype];
      pybind11::array_t<double> cl_array = pybind11::cast<pybind11::array_t<double>>(cl_array_obj);

      // Create the vector to return
      std::vector<double> result = cast_np_to_std(cl_array);

      // cl = 0 for l = 0,1
      result.at(0) = 0.;
      result.at(1) = 0.;

      return result;
    }

    // returns angular diameter distance for given redshift
    double class_get_Da(double z)
    {
      double Da = cosmo.attr("angular_distance")(z).cast<double>();
      return Da;
    }

    // returns luminosity diameter distance for given redshift
    double class_get_Dl(double z)
    {
      double Dl = cosmo.attr("luminosity_distance")(z).cast<double>();
      return Dl;
    }

    // returns scale_independent_growth_factor D(z) for CDM perturbations for a given
    // redshift (quantity defined by CLASS as index_bg_D in the background module)
    double class_get_scale_independent_growth_factor(double z)
    {
      double growth_fact = cosmo.attr("scale_independent_growth_factor")(z).cast<double>();
      return growth_fact;
    }

    // returns scale_independent_growth_factor_f f(z)=d ln D / d ln a for CDM perturbations
    // for given redshift (quantity defined by CLASS as index_bg_f in the background module)
    double class_get_scale_independent_growth_factor_f(double z)
    {
      double growth_fact_f = cosmo.attr("scale_independent_growth_factor_f")(z).cast<double>();
      return growth_fact_f;
    }

    // returns Hubble parameter for given redshift
    double class_get_Hz(double z)
    {
      double H_z = cosmo.attr("Hubble")(z).cast<double>();
      return H_z;
    }

    // returns Omega radiation today
    double class_get_Omega0_r()
    {
      double Omega0_r = cosmo.attr("Omega_r")().cast<double>();
      return Omega0_r;
    }

    // returns Omega of ultra-relativistic species today
    double class_get_Omega0_ur()
    {
      double Omega0_ur = cosmo.attr("Omega_ur")().cast<double>();
      return Omega0_ur;
    }

    // returns Omega matter today
    double class_get_Omega0_m()
    {
      double Omega0_m = cosmo.attr("Omega_m")().cast<double>();
      return Omega0_m;
    }

    // returns Omega ncdm today (contains contributions of all ncdm species)
    double class_get_Omega0_ncdm_tot()
    {
      double Omega0_ncdm = cosmo.attr("Omega_ncdm_tot")().cast<double>();
      return Omega0_ncdm;
    }

    // returns Omega_Lambda
    double class_get_Omega0_Lambda()
    {
      double Omega0_Lambda = cosmo.attr("Omega_Lambda")().cast<double>();
      return Omega0_Lambda;
    }

    // returns sound horizon at drag epoch
    double class_get_rs()
    {
      double rs_d = cosmo.attr("rs_drag")().cast<double>();
      return rs_d;
    }

    // returns optical depth at reionisation
    double class_get_tau_reio()
    {
      double rs_d = cosmo.attr("tau_reio")().cast<double>();
      return rs_d;
    }

    // returns redshift of reionisation
    double class_get_z_reio()
    {
      double rs_d = cosmo.attr("z_reio")().cast<double>();
      return rs_d;
    }

    // returns sigma8 at z = 0
    // (root mean square fluctuations density fluctuations within
    // spheres of radius 8/h Mpc)
    double class_get_sigma8()
    {
      double sigma8 = cosmo.attr("sigma8")().cast<double>();
      return sigma8;
    }

    // returns Neff
    double class_get_Neff()
    {
      double Neff = cosmo.attr("Neff")().cast<double>();
      return Neff;
    }

    // Returns Hubble parameter today
    double class_get_H0()
    {
      double H0 = cosmo.attr("Hubble")(0).cast<double>();
      return H0;
    }

    // print primordial power spectrum for consistency check & debug purposes
    void print_pps()
    {
      std::cout<< "Primordial spectrum from classy: "<< std::string(pybind11::str(cosmo.attr("get_primordial")())) << std::endl;
    }

  }
  END_BE_NAMESPACE

#endif

BE_INI_FUNCTION
{

  #ifdef HAVE_PYBIND11

    static int error_counter = 0;
    static int max_errors = 100;

    // get input for CLASS run set by CosmoBit
    Classy_input input_container = *Dep::classy_input_params;
    pybind11::dict cosmo_input_dict = input_container.get_input_dict();

    // check if any energy injection-related parameters are contained in the
    // CLASS input dictionary. If so, throw a fatal error as these models
    // requires features only available in exoCLASS
    // Note: checking for presence of input key 'f_eff_type' as this
    // input has to be added for all models making use of the energy
    // injection features (regardless of whether they are annihilating
    // or decaying DM models).
    if (cosmo_input_dict.attr("__contains__")("f_eff_type").cast<bool>())
    {
      std::ostringstream error;
      error << "\nYou are scanning over a model, e.g. decaying or annihilating DM, ";
      error << "\nwith exotic energy injection in the early Universe. However, you ";
      error << "\nare using a CLASS version that does not support these calculations. ";
      error << "\nGAMBIT will exit now. To fix this simply run";
      error << "\n\ncd build; make nuke-classy_"<< STRINGIFY(VERSION) <<"; make classy_exo_2.7.2; cd ..";
      backend_error().raise(LOCAL_INFO,error.str());
    }

    static bool first_run = true;

    if(first_run)
    {
      max_errors = runOptions->getValueOrDef<int>(100,"max_errors");
      cosmo = classy.attr("Class")();
      // check input for consistency
      class_input_consistency_checks(cosmo_input_dict);

      // create copy of cosmo_input_dict to cache results from
      // previous run
      prev_input_dict = cosmo_input_dict.attr("copy")();
    }

    // test if input arguments for CLASS are exactly the same as in previous run ...
    bool equal = compare_dicts(prev_input_dict, cosmo_input_dict);

    // .. if so there is no need to recompute the results. If not, clean structure, re-fill input & re-compute
    if(not equal or first_run)
    {
      // Clean CLASS (the equivalent of the struct_free() in the `main` of CLASS -- don't want a memory leak, do we
      cosmo.attr("struct_cleanup")();

      // Actually only strictly necessary when cosmology is changed completely between two different runs
      // but just to make sure nothing's going wrong do it anyways..
      cosmo.attr("empty")();

      // set cosmological parameters
      logger() << LogTags::debug << "[classy_"<< STRINGIFY(VERSION) <<"] These are the inputs:"<<endl;
      logger() << pybind11::repr(cosmo_input_dict) << EOM;
      cosmo.attr("set")(cosmo_input_dict);

      // CLASS re-computed -> save this information in cosmo container, so MontePython
      // (and potentially other backends) has access to this information
      cosmo.attr("set_cosmo_update")(true);
      // -> access value
      //int recomputed = cosmo.attr("recomputed").cast<int>();

      // Try to run class and catch potential errors
      logger() << LogTags::info << "[classy_"<< STRINGIFY(VERSION) <<"] Start to run \"cosmo.compute\"" << EOM;
      try
      {
        // Try to run classy
        cosmo.attr("compute")();
        // reset counter when no exception is thrown.
        error_counter = 0;
        logger() << LogTags::info << "[classy_"<< STRINGIFY(VERSION) <<"] \"cosmo.compute\" was successful" << EOM;
      }
      catch (std::exception &e)
      {
        std::ostringstream errMssg;
        errMssg << "Could not successfully execute cosmo.compute() in classy_"<< STRINGIFY(VERSION)<<"\n";
        std::string rawErrMessage(e.what());
        // If the error is a CosmoSevereError raise an backend_error ...
        if (rawErrMessage.find("CosmoSevereError") != std::string::npos)
        {
          errMssg << "Caught a \'CosmoSevereError\':"<<endl;
          errMssg << rawErrMessage;
          backend_error().raise(LOCAL_INFO,errMssg.str());
        }
        // .. but if it is 'only' a CosmoComputationError, invalidate the parameter point
        // and raise a backend_warning.
        // In case this happens "max_errors" times in a row, raise a backend_error
        // instead, since it probably points to some issue with the inputs
        else if (rawErrMessage.find("CosmoComputationError") != std::string::npos)
        {
          ++error_counter;
          errMssg << "Caught a \'CosmoComputationError\':"<<endl;
          errMssg << rawErrMessage;
          if ( max_errors < 0 || error_counter <= max_errors )
          {
            backend_warning().raise(LOCAL_INFO,errMssg.str());
            invalid_point().raise(errMssg.str());
          }
          else
          {
            errMssg << "\nThis happens now for the " << error_counter << "-th time ";
            errMssg << "in a row. There is probably something wrong with your inputs.";
            backend_error().raise(LOCAL_INFO,errMssg.str());
          }
        }
        // any other error (which shouldn't occur) gets also caught as invalid point.
        else
        {
          errMssg << "Caught an unspecified error:"<<endl;
          errMssg << rawErrMessage;
          cout << "An unspecified error occurred during compute() in classy_"<< STRINGIFY(VERSION) <<":\n";
          cout << rawErrMessage;
          cout << "\n(This point gets invalidated) " << endl;
          invalid_point().raise(errMssg.str());
        }
      }
      //std::cout << "Trying to print power spectrum..." << std::endl;
      //print_pps();

    }
    // identical CLASS input -- skip compute step & save time!
    else
    {
      logger() << LogTags::info << "[classy_"<< STRINGIFY(VERSION) <<"] \"cosmo.compute\" was skipped, input was identical to previously computed point" << EOM;
      // CLASS did not recompute -> save this information in cosmo container, so MontePython
      // (and potentially other backends) has access to this information and can skip
      // their computations as well
      cosmo.attr("set_cosmo_update")(false);
      // -> access the information with
      //int recomputed = cosmo.attr("recomputed").cast<int>();
    }

    first_run = false;
    // save input arguments from this run to dictionary prev_input_dict
    // (clear entries before copying)
    prev_input_dict.attr("clear")();
    prev_input_dict = cosmo_input_dict.attr("copy")();

  #endif

}
END_BE_INI_FUNCTION

Updated on 2023-06-26 at 21:36:57 +0000