How to customize the performance function (mse) to RMSE in AAN?

How to customize the performance function (mse) (net.performFcn = 'mse') in AAN? For example to chane mse to RMSE.

Thank you.

Answer 1

[TED MOSBY]

Hi Kwaku,

To change the performance function, you must implement a custom performance function to evaluate the performance as “net.performFcn”does not provide option for custom function . Thus, you cannot directly integrate it as “net.performFcn” expects the functions to be predefined . Hence you can evaluate the performance using your own custom function as shown below:

function rmse = calculateRMSE(targets, outputs)

mse = mean((targets - outputs).^2, 'all'); % Calculate MSE

rmse = sqrt(mse); % Calculate RMSE from MSE

end

outputs = net(inputs); % Get the network's outputs for the input data

rmse = calculateRMSE(targets, outputs); % Calculate RMSE using your custom function

For more information on net.performFcn refer below:

https://in.mathworks.com/help/deeplearning/ug/neural-network-object-properties.html?s_tid=srchtitle_site_search_1_net.performfcn

Hope this helps!

Best regards,

Kalash