I am having a doubt while deciding the batch size for both the training and validation datasets!
How should I determine if it should be 20/32/50/128,etc?
Try the different values and see which one works best.
Then there’s the famous quote from Yann LeCun: “Friends don’t let friends use batch sizes greater than 32.” Here’s the link to the Twitter post with the full quote.
1 Like
Haha… witty of you!
Well…Thank you so much!
Note that the Twitter thread included a link to a paper by LeCun’s group showing the results of a lot of experiments and concluding that over a pretty wide range of deep learning training tasks, batch sizes between m = 2 and m = 32 consistently gave the best performance.
So at least that limits your search space.
1 Like