After posting pull request #1, I subsequently accidentally pushed a bunch of experimental changes to that same branch which then got merged in. These included changes to the pickled file format to include the (computed) alphabet, and an experiment right padding all strings to the same length. These were not backwards compatible changes and the experiment with that code did not give me a model that converted.
So this pull request reverts the earlier one, but then re-adds the stable changes from earlier pull requests #1 and #3.
This version of the code is also compatible with a TF1.0 model I've uploaded here. Here's a sample session on how to use it:
This should result in several files in the tf10_model_figures subdirectory. Examples attached.
Sorry if these experimental changes caused any confusion - happy to discuss that code in case there are changes you would like to consider merging in. But for now I think it's best to revert to this stable version which has a trained model.
After posting pull request #1, I subsequently accidentally pushed a bunch of experimental changes to that same branch which then got merged in. These included changes to the pickled file format to include the (computed) alphabet, and an experiment right padding all strings to the same length. These were not backwards compatible changes and the experiment with that code did not give me a model that converted.
So this pull request reverts the earlier one, but then re-adds the stable changes from earlier pull requests #1 and #3.
This version of the code is also compatible with a TF1.0 model I've uploaded here. Here's a sample session on how to use it:
This should result in several files in the
tf10_model_figures
subdirectory. Examples attached.Sorry if these experimental changes caused any confusion - happy to discuss that code in case there are changes you would like to consider merging in. But for now I think it's best to revert to this stable version which has a trained model.