Jose Flores
Sep 8, 2018 · 1 min read

Hi Felipe, thanks! I originally planned for this to be done as a Jupyter notebook but I wanted to focus on saving and loading and I worried that if I included a more full example people could get lost in the specific details of the model being trained.

I think you’re right though and I should have included at least a small/full project that shows saving and loading.

  • The tags are just identifiers and if they’re aren’t provided then the tensors would have some default value.
  • This doesn’t exactly save checkpoints but it can give you similar functionality. In the article I talked about it briefly in the “How is this different from tf.train.saver” section and the gist is that checkpoints will save your data but not your graph structure so if you used checkpoints then you will have to have the graph reconstructed. Using the SavedModel api to save your model will save your data and your graph structure.
  • Thanks for catching that! it should be trying to run the myOutput tensor and it feeds in the myInput tensor.
    Jose Flores

    Written by

    A passionate Software Engineer with a focus in Android development and a love for solving challenging problems.