When you run the function nnet of the nnet package a sequence of values is shown on the console like this (made up numbers):

initial value 100

iter 10 value 88

iter 20 value 80

final value 60

And it shows "Converged" at the end if the net did converge.

According to the documentation, this value is the "value of fitting criterion plus weight decay term". I know what the decay term is but not the fitting criterion. I tried looking for the exact meaning on the web but I didn't find anything useful.

Could you explain to me the meaning of these values? Do they say something about the convergence of the net? Should I look at them?

  • $\begingroup$ fitting criterion is error (loss) between ground truth and network output. The numbers are values of objective function after iterative minimization. The value of objective function (fitting criterion + weight decay) gets smaller at each iteration since weights updated in a direction which minimizes objective. Probably, you see that after some iterations the values converge as objective function gets closer to a local minima. Check for MSE or cross entropy function rather than fitting criterion. $\endgroup$ Jun 10, 2015 at 21:35

1 Answer 1


The fitting criterion is another synonym for loss function. Training a neural network proceeds by adjusting the weights and biases of a neural network to minimize the loss function (fitting criterion) with respect to the training data.

See also: Objective function, cost function, loss function: are they the same thing?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.