I’ve seen several different places say that, when an RNN and especially if you’re using attention, if you train first on shorter sentences and later on longer sentences, you’ll get better results.

I understand why, but how do people execute this in practice? I’m fairly certain going strictly -> in the exact order is not a great idea, so I’d love to be able to get an almost exact to long ordering of my dataset…. is there an easy function or in python that does this that I’ve overlooked? If not, how do people do it otherwise?

Source link
thanks you RSS link
( https://www.reddit.com/r//comments/8rw0a4/d__training_from_short_to_long/)


Please enter your comment!
Please enter your name here