Transference of weights leading to smaller model (Keras)

Published: 12/06/2019
Transference of weights leading to smaller model (Keras)
Source: STACKOVERFLOW.COM

I am trying to insert a dropout layer into my model. So I load the old model, create a new model architecture, and transfer the weights. However when I save the new model, the memory footprint is a lot smaller: 113 megabytes vs original 338 megabytes. I suspect I must be making a mistake in the process, but the model seems to save and is running. The accuracy is about 15% lower but can't tell if

Read more
Related news
Comment
FACEBOOK