How to re-load MatLab's feedforwardnet to Workspace?

I create a feed-forward neural network 100 times, each time with different input and target. The trained model is saved into "FFNN_trained_sum" each time.

FFNN_trained_sum = []
for iteration = 1:100;
    FFNN = feedforwardnet(20);
    x = ...; t = ...; % assign the input "x" and target "t" with different datasets. 
    FFNN_trained = train(FFNN,x,t);
    FFNN_trained_sum = [FFNN_trained_sum, FFNN_trained]
end

So after 100 times, the "FFNN_trained_sum" is a 1x100 network. How can I load the 37th network into MatLab Workspace? I tried the below but got error.

>> FFNN_trained_sum(37)
Scalar structure required for this assignment.

Error in network/sim (line 141)
net.trainFcn = ''; % Disable training related setup

Error in network/subsref (line 15)
        otherwise, v = sim(vin,subs{:});

Answer 1

[Jon Cherrie]

I recommend using a cell array rather than a regular array for this case, e.g.,

FFNN_trained_sum = {}; % Use {} here, not []
for iteration = 1:100;
    FFNN = feedforwardnet(20);
    x = ...; 
    t = ...; 
    FFNN_trained = train(FFNN,x,t);
    FFNN_trained_sum{end+1} = FFNN_trained; % append the new network at the end
end

You can then access the 37th network via FFNN_trained_sum{37}

[Cheng Zhang]

Sure, is there a way to convert the regular array to a cell array?

[Jon Cherrie]

If you already have that array of networks, and want the 37th one, you could try this:

s = struct(FFNN_trained_sum);
FFNN_trained_37 = network(s(37));

To convert the array to a cell array, this seems to work:

cellOfNetworks = {};
s = struct(FFNN_trained_sum);
for i = 1:length(s)
    cellOfNetworks{end+1} = network(s(i));
end