cambridge-iccs / fortran-tf-lib Goto Github PK

Need to include some examples which replicate those from the PyTorch library. This will need to include some with Keras SavedModel which is the Torch Script equivalent.

I suggest adding topics such as fortran, tensorflow, pytorch, keras, machine-learning, interoperability in the About section at https://github.com/Cambridge-ICCS/fortran-ml-bridge.

• Unit tests where possible.
• Integration tests.

It would be handy for conda users to be able to install from a channel.

Stretch goal: Add conda build to CI

Documentation can be duplicated from PyTorch library and placed in README. Needs some specific input from @SimonClifford on TensorFlow.

There does seem a bit of an asymmetry in the PyTorch examples where the C, C++ and Python examples take user arguments for the model file, but the Fortran has the model filename is hard-coded. It's not a big deal but maybe we should consider make the Fortran take command line arguments too?

Running make install installs the project to the current CMAKE_INSTALL_PREFIX set of directories, however the generated
CMakeLists.txt file does not correctly set up the project for subsequent use with find_package(FortranTensorFlow). What should be possible should be something like:

find_package(FortranTensorFlow)
target_link_libraries( foo FortranTensorFlow::fortran-tf)

where cmake will determine where the module files from the package. Currently this does not work, and the package does not set any include directories.

This should include general user documentation rendered to readthedocs via Sphinx. There is also an option to include the API docs within this site (doxygen).

• Create a sensible Readme.md following the guidance in our handbook (development guide)
• Create walkthrough examples in the documentation.

Update (03/02/2023)

• TODO