NEWS.rst

New in 0.8:

• added the possibility to stop remaining iterations based on the reached level of precision
• WARNING: removed global callback functions; each integrator now takes an optional fourth argument which must be the callback function. The default callback function is the verbose one. Also the function names of the previously called _default_callback functions to _silent_callback
• WARNING: changed the interfaces of the integrators hep::plain, hep::vegas, and hep::multi_channel in order to support the new checkpointing system (see below)
• added checkpointing system: It is now possible to seamlessly resume an integration starting from a previously generated checkpoint. Checkpoints are the new return type of every integrator and can be easily saved to and read from text files; see examples/checkpoints.cpp
• removed Google Test and replaced it by Catch2; this dependency should be downloaded automatically in the meson configuration step if it isn't present already
• WARNING: changed the interface of hep::plain and hep::mpi_plain to perform multiple iterations rather than a single big one. Between two iterations a callback function is invoked, like with any other algorithm as well
• WARNING: at this point the autotools-based build system is longer supported. Use the meson build system instead
• added support for two-dimensional distributions
• WARNING: the member function mid_points is now removed and replaced by the two free-standing functions mid_points_x and mid_points_y
• WARNING: changed some member functions of the classes hep::distribution_parameters and hep::distribution_result. The changes were neccesary to support two-dimensional distributions.

New in 0.7:

• WARNING: changed the multi channel map function interface. This function must now also accept a vector of enabled channels

New in 0.6:

• added functionality to accumulate the results of distributions together with the integrated results
• changed the interfaces of functions that return accumulated results
• WARNING: removed the parameters() function in hep::projector
• added the possibility to use Meson for building/installing hep-mc

New in 0.5:

• fixed a bug that caused NaNs to slip through to the integrated result if they were coming from the point's weight
• it is now possible to disable channels by setting their weight to zero
• the function multi_channel_verbose_callback now prints an overview of the smallest and largest a-priori weights
• the functions multi_channel and mpi_multi_channel now accept an optional parameter min_calls_per_channel that limits the a-priori weights to a smallest weight determined by a minimum number of calls. If set to zero (the default), this corresponds to the old behavior
• WARNING: this version introduces many interface changes; if you used previous version of this library you must update your code accordingly
• implemented lazy evaluation of the weight for the multi channel integrator. This required to change the interface of the map function which now must accept an additional parameter action that determines whether the coordinates or the densities must be calculated. This in turn enables the algorithm to skip the evaluation of densities when the integrand returns a zero value
• changed name of density_function() in hep::multi_channel_point to map()
• when calling the integration algorithms the numeric type must no longer be explicitly specified. The type information is now infered from the integrand
• renamed the class vegas_iteration_result to vegas_result
• completely changed the interfaces of all integrators; the functions that are integrated must now be specified by using an hep::integrand or hep::multi_channel_integrand which are obtained by using hep::make_integrand or hep::make_multi_channel_integrand. This allows the library to support the generation of differential distributions for all integrators. If distributions should be generated, the integrands must accept a second parameter, a reference to hep::projector, that takes care of the binning. The results of the distributions are captured in a hep::distribution_result for each distribution
• hep::plain now returns a hep::plain_result which contains the results for possibly generated distributions
• all _result classes except hep::mc_result derive now from hep::plain_result
• encapsulated all members of the class mc_point and classes deriving from it
• WARNING: interface changes to the class mc_point and all deriving classes; the member function does no longer include the averaging factor 1/N where N is the number of calls. If you used this member variable before, you MUST CHANGE YOUR CODE accordingly

New in 0.4:

• added multi channel MC integrator with example
• added a new MC helper function hep::multi_channel_max_difference that calculates the maximum difference for adaptive optimization in multi channel MC
• added callbacks for multi channel MC analogous to the case for VEGAS
• added MPI parallelized routines for multi channel MC

New in 0.3.1:

• added new internal class hep::kahan_accumulator. This class takes care of the kahan summation
• changed interfaces mc_result and vegas_iteration_result: Their member variables are now properly encapsulated and accessible via public member functions. Furthermore, the mc_result now allows to obtain sum(), sum_of_squares(), and variance()

New in 0.3:

• fixed compilation error
• removal of the hep::mpi_single_generator function. This is now on by default - the integration results are independent of the number of processes
• temporary fix for single-precision floating point numbers that are 1.0f and therefore outside of the half-open interval [0,1)
• changed hep::vegas_pdf interface to use more descriptive member function names and moved the member function icdf outside the class
• revamped documentation
• changed function names of hep::cumulative_result and hep::chi_square_dof to the same names with a 0 and 1 at the end of their names; 0 uses the same algorithm as before, 1 weighs all results equally. The function hep::cumulative_result remains but requires an additional argument
• added new global configuration function hep::vegas_cuba_refinement() that can be used to modify pdf refinement to use CUBA's method
• removed --enable-tests configure switch. All external dependecies are gone and running make check is sufficient now to run tests
• changed interfaces hep::cumulative_result and hep::chi_square_dof which no longer need to be called with the numeric type; the type is automatically determined by the iterators
• support for MPI can now be enabled with --enable-mpi independently from the examples. If MPI is enabled, the MPI examples are built if examples are activated and the MPI tests are checked if tests are activated. The MPI headers are installed in any case because MPI must be explicity requested by including hep/mc-mpi.hpp
• renamed hep::linear_grid to hep::vegas_pdf and moved code from VEGAS into this class
• add more unit tests, use Google Test as testing framework
• improved the MPI VEGAS example
• added new hep::mpi_single_generator function that adds the possibility to use the same random numbers for MPI and non-MPI routines. This makes hep::mpi_vegas return the same numerical result as hep::vegas
• fixed integer overflow bug on 32-bit architectures that caused VEGAS' grid adjustment to fail when the number of calls surpassed 2^16
• modified VEGAS callback functions to return a boolean variable signaling to stop all remaining iterations
• modified the verbose callback function to print the error in percentage

New in 0.2.2:

• added two examples showing the advanced use of the VEGAS integration routine
• added a callback mechanism for the VEGAS routines; this enables one to print intermediate results from completed iterations
• added VEGAS routines that reuse existing grids
• grids can now be easily saved to files with iostream shift operators, see the example examples/read_linear_grid.cpp which can be used with the grid file in examples/grid_file

New in 0.2.1:

• added MPI-parallelized PLAIN integration algorithm
• lowered dependency on autotools features

New in 0.2:

• added new functions for combining several VEGAS results
• added MPI-based paralellized VEGAS integration algorithm
• improved documentation
• improved VEGAS performance

New in 0.1:

• added PLAIN integrator
• added VEGAS integrator