OpenNMT Forum

Familiarity with the keywords in the log

Hello , I am using OPenNMT for my research purpose . In our paper ,we would like to make a comparison where we need to know about every keywords we will be talking about . I would like to know about following words and their functionality :
**acc (I guess its the accuracy of training set…if i am wrong please correct me )
**ppl (I dont know about it)
**xent (I dont know about it)
**lr (I guess its the rate how fast or slow my model is getting learnt …if i am wrong please correct me)
**validation accuracy (I guess its the accuracy of validation set…if i am wrong please correct me )
***validation perplexity (I dont know about it)

and another question : I am using warmup steps same as my training steps for transformer model . Will it make any difference or I would like to know will my model be accepted??

I am sorry if I ask some silly questions .Thanks Beforehand

Best ,
Argha Dhar

Is there anyone ?? May I please get any reply??

I’m afraid no one here will give you a full course about all this. Most of the answers you’re looking for are already out there. You might want to do some research, e.g. read some introductory blog posts or papers to machine learning, deep learning and neural machine translation if these topics are unknown to you.

acc = accuracy, basically how good is the model at generating the next tokens, knowing the previous ones
ppl = perplexity
xent = cross entropy
These are training metrics, defined here https://github.com/OpenNMT/OpenNMT-py/blob/073428849c1d10dd4fae7f8fd92699cdc9f230a4/onmt/utils/statistics.py#L10

lr = learning rate, you definitely should know about this one if diving into a deep learning project

1 Like