This is a wrapper class for a slightly modified cnpy, which can write std::tuple objects developed for use in Geant4 simulations. While it was developed for the framework of Geant4, but it may be useful for other threaded applications, as such there is no dependence on Geant libraries.
This updated cnpy code can be found within the forked repository Lala5th/cnpy.
Requires libizp to function.
To build you will need cmake. To install do
mkdir build
cd build
cmake -DCMAKE_INSTALL_PREFIX=/path/to/install/dir ..
make
make installAfter install it should work with cmake's find_package() command as well as with using
-L/path/to/install/dir -lG4NumpyAnalysis
with g++.
To use it include the NumpyAnalysisManager.hh header file.
It implements a proprietary analysis manager, which is quite different from the analysis managers provided by Geant4. It is accessed similarly via invoking NumpyAnalysisManager::GetInstance(bool continousWrite = true), which returns the active instance or creates an instance if it's not yet initialised. If it is not yet initialised the boolean parameter decides whether the data will be written at once when WriteData() is called, or continously into separate .npy files and only merged into an npz archive at the time of the WriteData() call. The other methods are SetFilename(std::string), which changes the name of the output npz archive (default is out.npz). CreateDataset<typename... COLS>(std::string title), which creates a new datatable with COLS as columns, while returning the id of the new datatable. If the manager is in continous write mode this also creates a new datafile named title.npy. AddData<typename... COLS>(int id, COLS...) adds a row to the datatable with the matching id [!Not checked in not continous write, but same template arguments must be used, or template can sometimes be dropped in second case. Care must be taken as it only works in specific cases where it can guess type specifically, otherwise leads to unwanted/undefined behaviour]. Finally WriteData() writes the npz object to disk.
While it is technically thread-safe, it is advisable that worker threads do not access it, otherwise (if badly handled), the datatables might store the same data multiple times. In Geant4 the IsMaster() function can be quite helpful so worker threads do not accidentally smear the data.
- Write Basic manager
- Add mode for writing invidual datafiles
- Add way to group several columns together (i.e. group vectors into a special dtype)