Reading CHECKPOINT

[ArvinZJC]

I saw issue #144 . However, I still get a NotFoundError(/data/user/0/com.arvinzjc.xshielder/files/chaquopy/AssetFinder/app/cnn_trainer_saver; No such file or directory) in my project.

The relevant files are:
src/main/python/code.py
src/main/python/cnn_trainer_saver/cnn_trainer.ckpt

The relevant code is:
tf.train.Saver().restore(session, os.path.join(os.path.dirname(__file__), 'cnn_trainer_saver', 'cnn_trainer.ckpt'))

[mhsmith]

Are you sure you're using this API correctly? According to the Saver documentation, the path parameter is not a complete filename but a "path prefix of the newly created checkpoint files".

[ArvinZJC]

More specifically, the relevant files are:
src/main/python/code.py
src/main/python/cnn_trainer_saver/cnn_trainer.ckpt.meta
src/main/python/cnn_trainer_saver/cnn_trainer.ckpt.index
src/main/python/cnn_trainer_saver/cnn_trainer.ckpt..data-00000-of-00001
src/main/python/cnn_trainer_saver/checkpoint

The exception occurred when I put these 4 checkpoint files in the folder cnn_trainer_saver. If I remove 'cnn_trainer_saver' from the code and move the files to the same directory of the PY file (i.e. src/main/python/(these checkpoint files)), then everything looks good. My question is actually why this happens. Thanks!

[mhsmith]

Unless this behaviour is different on Chaquopy than it is on other platforms, I think this is a question about TensorFlow rather than Chaquopy. I'm not an expert on TensorFlow, so you should ask on the TensorFlow GitHub page, or on StackOverflow.