This PR adds initial support for generating a Tensorboard log directory when doing the training step. The only metric supported for now is Average_Error, but it should be easy to add more in the future. Adding new kinds of metrics (i.e. not only scalars) should also be possible by using the Tensorflow C++ API.
As part of this change, it becomes necessary to build and install Tensorflow from scratch so that we get access to the C++ API. Doing that is a giant pain in the neck and adds a lot of time to the initial setup. I'm open to suggestions on how to make this better.
Also, this PR is built on top of #173 because using CMake makes it much easier to take on new dependencies.
This PR adds initial support for generating a Tensorboard log directory when doing the training step. The only metric supported for now is
Average_Error, but it should be easy to add more in the future. Adding new kinds of metrics (i.e. not only scalars) should also be possible by using the Tensorflow C++ API.As part of this change, it becomes necessary to build and install Tensorflow from scratch so that we get access to the C++ API. Doing that is a giant pain in the neck and adds a lot of time to the initial setup. I'm open to suggestions on how to make this better.
Also, this PR is built on top of #173 because using CMake makes it much easier to take on new dependencies.