jiesutd / NCRFpp

NCRF++, a Neural Sequence Labeling Toolkit. Easy use to any sequence labeling tasks (e.g. NER, POS, Segmentation). It includes character LSTM/CNN, word LSTM/CNN and softmax/CRF components.
Apache License 2.0
1.89k stars 446 forks source link

Why is test_dir included in the training config? #141

Closed taasmoe closed 5 years ago

taasmoe commented 5 years ago

I've noticed some earlier issues concerning the use of the test set for initializing vocabularies.

Even if it won't affect the performance, why should we monitor the test results during training?

Thanks

jiesutd commented 5 years ago

Because in the beginning, I didn't implement the model saving module, so monitor test result is necessary to report the experiment results.

But as now the model saving module has been implemented, it is your personal choice to decide whether to monitor the test data during training. For me, I keep it because I am lazy and I don't want to add an extra step of loading the model and decode test data.

I have checked and confirmed it does not affect the model performance. Just for convenience.