tierra-colada / h5geo Goto Github PK

C++ and python API to work with geo-data based on HDF5 containers through h5gt header only library.

For now h5geo provides API to work with:

• seismic data
• wells (deviations, logs, welltops)
• maps
• points
• horizons

h5geo also includes h5geopy project wich is a python wrapper.

h5geopy uses pybind11 to make python bindings and there may be some perfomance penalty when copying matrices from C++ Eigen3 matrices to numpy (or std::vector to py::list). I have not done any perfomance testing.

This project is aimed at geoscientists and developers

Common documentation: h5geo.readthedocs.io

C++ API documentation: tierra-colada.github.io/h5geo

h5geo depends on:

• HDF5
• h5gt (header only)
• Eigen3 (header only)
• magic-enum (header only)
• units
• mio (header only)
• optionally: TBB
• optionally: GDAL
• optionally: h5gtpy (needed only when building with python support -DH5GEO_BUILD_h5geopy=ON)

You can use superbuild or build every dependency by yourself. Then do:

mkdir build
cd build
cmake ..
  -DCMAKE_BUILD_TYPE=Release
  -DEigen3_ROOT:PATH=/path/to/eigen 
  -Dmagic_enum_ROOT:PATH=/path/to/magic_enum 
  -Dmio_ROOT:PATH=/path/to/mio
  -Dh5gt_ROOT:PATH=/path/to/h5gt 
  -DH5GEO_SUPERBUILD:BOOL=OFF
  -DH5GEO_USE_THREADS:BOOL=ON
  -DTBB_ROOT:PATH=/path/to/tbb
  -DH5GEO_BUILD_SHARED_LIBS:BOOL=ON
  -DH5GEO_BUILD_h5geopy:BOOL=ON
  -DHDF5_RUNTIME_DIR:PATH=/path/to/dir/containing/hdf5-runtime
  -DH5GEOPY_COPY_RUNTIME_DEPS:BOOL=ON
  -DPYTHON_EXECUTABLE=/path/to/python.exe 
  -DH5GEO_BUILD_TESTS:BOOL=OFF
  -DCMAKE_INSTALL_PREFIX:PATH=/install/prefix
  -G "some generator"
cmake --build . 

NOTE: instead of <prefix>_ROOT path (root directory of every package) you can pass <prefix>_DIR path containing some <prefix>-config.cmake file of every package.

NOTE: to run h5geopy python must know where HDF5-runtime, units-runtime, tbb-runtime and h5geo-runtime are located. You may choose whether to copy them to site-packages/h5geopy dir or add their dirs to PATH env. By default cmake try to search h5geo-runtime in h5geo-install/lib folder. Also h5geopy depends on h5gtpy. So you would need to have installed h5gtpy.

h5geo installation is done the same way one usually installs cmake project:

cmake --install . --prefix /some/path --config Release

Then you can find h5geopy in site-packages of your python env.

NOTE: cmake tries to find runtime dependencies at install step. If there are unresolved or conflict dependencies then you need to modify PATH env, rerun cmake .. and cmake --install . --prefix /some/path --config Release. You don't have to rebuild the project!

It is possible to build and install the library using the following command:

pip install git+https://github.com/tierra-colada/[email protected] --verbose

This will automatically run CMake Superbuild to build build h5geo v0.1.0 and its dependencies (without GDAL support) in the site-pakages/h5geopy dir.

On modern PC this usually takes a couple of minutes.

NOTE: CMake and C++ compiler are needed.

Importing is done via:

from h5gtpy import h5gt
from h5geopy import h5geo

You can find tests in site-packages/h5geopy. To run them:

python -m unittest discover h5geopy.tests

Note: h5geopy (as well as h5geo) uses fortran memory order (columnwise). Thus you should allocate numpy with numpy.asfortranarray or order='F' as numpy uses C order by default.

There is stub file h5geopy/h5geo.pyi. You need to set VS Code setting "python.analysis.stubPath": "some/relative/path/to/stub-storage": where stub-storage is supposed to contain folder h5geopy with h5geo.pyi file.

I'm some kind of a uncomfortable to distribute stub file, so to be sure that stub file is updated you can generate your own stub file:

pip install pybind11-stubgen
pybind11-stubgen --no-setup-py h5geopy._h5geo

You don't need to provide stub file to PyCharm as it handles this by its own.

1.10, 1.12 (recommended)

Windows 10/11, Linux, OSX