import autoEncoder from workspace into Deep Network Designer APP

Hi everybody,

Using the Deep Learning toolbox, I've created and trained an AUTOENCODER from AUTOENCODER CLASS.

I can view the result using "view(autoenc)".

In order to "tweek" it, I've tried to transfer it to the "DEEP NETWORK DESIGNER APP" by using a conversion "net = network(autoenc)" => it does not work and I got a panel info that I cannot understand.

Is it possible to transfer the AUTOENCODER from workspace to the APP ?

If not how, is it possible to recover the exact definition of my AUTOENCODER (layers types, sizes,...) in order to recreate it in the APP ?

have a nice Day

Juliette

Answer 1

[David Willingham]

Hi,

The autoencoder function was built using MATLAB's neural network legacy functionality, I.e. prior to when we released an updated deep learning framework in R2018b. As such it won't worh with the Deep Network designer.

If you created an autoencoder using the updated framework, these can be imported into Deep Network Designer.

This autoencoder page lists examples that use the newer framework.

[juliette soula]

Hi David,

thanks for feedback, I got the "legacy" point and wont spend more time trying to transfer what is already achieved with the old function into the new framework.

let's start again from beginning

BR

juliette