First, are you sure you don't want to create a single live TensorBoard for all of your data? Remember, one TensorBoard can show more than one experiment, because it can show arbitrarily many "runs", and will show you charts comparing them, will let you turn them on and off, etc. You can do this by making a directory structure where each of your experiments is in a separate subdirectory.
If you really want a static TensorBoard, this is possible too, although it is not supported. To see an example, go to
tensorflow.org/tensorboard
Here is what you need to do:
- You need to transform the data from .tfevents. into json. There is a script that does this: tensorboard/scripts/serialize_tensorboard.py. (You will need to use Bazel to run it.)
- You'll want to package all the html, the simplest way to do this is to use
vulcanize with --inline-scripts, --inline-css, etc. (To get started, I would get the TensorBoard development environment set up, and the tensorboard/demo/index.html loaded so you can learn from it.)
FYI, I am making some changes to this code, specifically removing tf-tensorboard-demo.html and folding it into tf-tensorboard.html, so you may want to wait a day or two so the code doesn't move under you.