Bug in Fit Report for Nlarx Models in System Identification App

Hi

Using the System Identification App there is clearly a bug in reporting the fitting. In the Nlarx model it always returns the same fit values although they are not correct. And do not match the model NRMSE. Is this a known issue?

Answer 1

[Debraj Bhattacharjee]

Can you provide us the reproduction steps where you always get the same fit?

In addition, you see a different value of fit for 'Prediction' focus (as shown in the report) vs on the plot because of the difference between 'Simulation' and 'Prediction' focus. See below for more details:

https://www.mathworks.com/help/ident/ug/definition-simulation-and-prediction.html

[Rajiv Singh]

By default the estimation is performed with "prediction" focus and the response plot is computed for the "simulation" scenario. The fit numbers you see in the model display correspond to the 1-step prediction errors which is what is minimized for training the model. To train a model for simulation error minimization, do:

opt = nlarxOptions('Focus', 'simulation');

model = nlarx(data, <regressors>, opt)

Or, in the app, set the estimation focus to simulation: