Is it really working for EARLY STOPPING in deepspeech? how i will check?


(Murugan R) #1

in my training process i was enabled EARLY STOPPING=TRUE. but not any impact, when my validation loss goes to over-fitting.

thank you.


(Lissyx) #2

You need to share more informations, because on our side, it works.


(Murugan R) #3

i was trained my model, in that cross validation, validation loss same value repeated 3-4 times. and then validation loss increasing also. when cross validation if early stopping true, it will leave the remaining epoch. but it doesn’t happen.

thank you so much for your response sir


(Lissyx) #4

First, please avoid using screenshots. Keep the logs as text, otherwise it’s useless and painful for analysis.

Second, I don’t see any Validation step performed here, so how can you expect early stop to work?


(Murugan R) #5

i think i made some mistake.

Parameter --validation_step needs to be >0 for early stopping to work

in my code “–early_stop True --earlystop_nsteps 6 --estop_mean_thresh 0.1 --estop_std_thresh 0.1” not mention validation_step. defaultly it takes 0.

sorry sir. i will try again. and update my status