Prediction Outputs

[selcouthlyBlue]

I noticed that the output layer you used for your architecture is a ctc decoder. If I recall correctly, the outputs generated by these decoders is a SparseTensor. How do you get the predicted characters from there without using converting them into dense tensors through sparse_to_dense?

[hugrubsan]

This decoding is only implemented in "test.py" and is done word by word using a dictionary that contains the correspondences between labels (as integers) and characters.
Exactly in these lines:

dec=session.run(decoded[0],test_feed) 
output = str(list(map(dct.get, list(dec.values))))

[selcouthlyBlue]

I guess you just need the values property of the SparseTensor generated by the decoder and nothing else?

[hugrubsan]

Yes, I just need those values.