Hi @freesunshine0316 . As I said in another issue, I'm having the correct BLEU results after using the correct tokenizer. Thank you again.
I was testing the code in python 3 (with some small changes, just to run the code (e.g range/xrange)). Using python 3, I trained the model and I got very different results (unfortunately, very low bleu scores). I was wondering if you tested the code in Python 3 and could help me if some tip.
I trained the model a few times and I did not find why the results (bleu) were different from python2. I checked the data generated by G2S_data_stream using python 2 and 3, and they are the same (with no shuffle)!
Maybe, there is some issue in some part of the model that I was not able to find yet, like that.
I have been looking the code several times, but I could not find the issue yet. Maybe because I'm not a tensorflow expert.
Hi @freesunshine0316 . As I said in another issue, I'm having the correct BLEU results after using the correct tokenizer. Thank you again.
I was testing the code in python 3 (with some small changes, just to run the code (e.g range/xrange)). Using python 3, I trained the model and I got very different results (unfortunately, very low bleu scores). I was wondering if you tested the code in Python 3 and could help me if some tip.
I trained the model a few times and I did not find why the results (bleu) were different from python2. I checked the data generated by G2S_data_stream using python 2 and 3, and they are the same (with no shuffle)!
Maybe, there is some issue in some part of the model that I was not able to find yet, like that.
I have been looking the code several times, but I could not find the issue yet. Maybe because I'm not a tensorflow expert.