save a neural net in matlab 2017b, can't load it in 2018b

Hi, the ANN in question is just a simple FF ANN stored in a file. But when I try loading it into matlab 2018b using

load('JacobianHessian.mat')

I get the following error message:

Warning: Class 'network' is an unknown object class or does not have a valid 'loadobj' method. Element(s) of this class in array 'Net' have been converted to structures.

Any workaround?

Thanks a lot, Gidon Eshel, [e-mail], [phone]

"

[Walter Roberson]

How was the network created? Was it with newff?

[Gidon Eshel]

THANK YOU. THis is the relevant code segment that generates the ANNs:

     net = feedforwardnet([6 5 4]);
      net.divideFcn = '';
      net.trainParam.showWindow = false;
      net = train(net,X(k,:)',y(k)');

Thanks again, GIdon

Answer 1

[Steven Lord]

Do you have Deep Learning Toolbox installed in your installation of release R2018b? Use the ver function to check. I suspect you don't.

[Gidon Eshel]

 >> ver
-----------------------------------------------------------------------------------------------------
MATLAB Version: 9.5.0.944444 (R2018b)
MATLAB License Number: ******
Operating System: Mac OS X  Version: 10.13.6 Build: 17G65 
Java Version: Java 1.8.0_152-b16 with Oracle Corporation Java HotSpot(TM) 64-Bit Server VM mixed mode
-----------------------------------------------------------------------------------------------------
MATLAB                                                Version 9.5         (R2018b)
Simulink                                              Version 9.2         (R2018b)
Control System Toolbox                                Version 10.5        (R2018b)
DSP System Toolbox                                    Version 9.7         (R2018b)
Global Optimization Toolbox                           Version 4.0         (R2018b)
Image Processing Toolbox                              Version 10.3        (R2018b)
Instrument Control Toolbox                            Version 3.14        (R2018b)
MATLAB Compiler                                       Version 7.0         (R2018b)
Mapping Toolbox                                       Version 4.7         (R2018b)
Optimization Toolbox                                  Version 8.2         (R2018b)
Signal Processing Toolbox                             Version 8.1         (R2018b)
Simulink Control Design                               Version 5.2         (R2018b)
Statistics and Machine Learning Toolbox               Version 11.4        (R2018b)
Symbolic Math Toolbox                                 Version 8.2         (R2018b)
>>

[SL: Edited to remove license number]

Answer 2

[Gidon Eshel]

so as you see, I do have statss and ML toolbox installed.

Thanks!

Answer 3

[Gidon Eshel]

but, I am warming to your idea. Indeed, I do NOT have Deep Learning installed. Am on it right now, will report back if this does it. Thanks!

Answer 4

[Gidon Eshel]

solved. Yes, the problem was overlooked the need to install both Stats&ML AND Deep Learning toolboxes. Sorry.

Thanks to all, best, Gidon

[Steven Lord]

The feedforwardnet function is part of Deep Learning Toolbox, as you can see from the navigation frame on the left part of its documentation page. [In release R2017b it was part of Neural Network Toolbox; that toolbox was renamed in release R2018b.]

FYI according to the product requirements page Deep Learning Toolbox requires only MATLAB. It does recommend that you have six other products. [If you want to use GPU support in Deep Learning Toolbox, Parallel Computing Toolbox moves from being recommended to being required.]

Statistics and Machine Learning Toolbox is not one of the recommended products, so you don't need Statistics and Machine Learning Toolbox to use feedforwardnet. It may be required for another part of your application or project, though.

Answer 5

[Gidon Eshel]

Thank you. My research and data analysis definitely require the stats box, but luckily there is no conflict, I can have my deep learning and my stats too!

All the best, thanks for the diligence, Gidon