How do I get the BLEU scores from NMT training and why does "validation_out.txt" only contain .</S>

[kulma111]

Hey everybody,
I am aware, this repository is quite old and am not sure if it is still maintained, but I have to implement exactly this approach for a course at university.

I am working with a virtual environment set to Python 2.7 and all the necessary additions (at least the example runs) on a Ubuntu Subsystem (I have Windows).

I'm trying to run and understand the machine translation example with all the default settings and datasets provided here, and was wondering, why there are no epochs (default wise, there seem to be many, many iterations but I only epoch 0). Since I dont have access to a GPU yet, I am only doing about 50 iteration to try to debug and understand all the parameters. I was thinking, that I could compute and anlyse the BLEU scores and set the 'bleu_val_freq' to 49 (to include it in the 50 iterations I'm doing).
A "validation_out.txt" file is produced but it only contains . 63 times (why 63?). This doesnt change, even if I change the number of iterations but I will try to do it again with many more iterations once I get access to the GPU. The documentation of Blocks and Fuel sadly could not really inspire any explanations.

Also, dont get any log outputs (without changing anything on the code) and theano inconsistency errors (see below) which don't interrupt the code. In a previous projekt I had similar errors, and I read, it was safe to ignore them.

I would greatly appreciate any help or advice :)