# Need some more explanation about generative loss and discriminative loss

I want to understand what the output below mean?
Generator loss: 0.9319421716928482, discriminator loss: 0.6268734838962555

Does Generator loss mean Generator has learnt to fool the discriminator 93% ?

Does discriminator loss mean discriminator has learnt to catch the fakes correctly 62%

-swetha

Yes I think your understanding is pretty much right, they are against each other!

The loss values are just pure numbers. They do not express prediction accuracy. In most cases, the loss values are defined by a function that has non-negative values and all you know is that â€ślower is betterâ€ť, so your goal is to minimize them. Of course there are exceptions, like Wasserstein Loss in Week 3, where the function is not constrained to be non-negative. But typically the loss function is BCE loss (â€śbinary log lossâ€ť) which has output values between 0 and \infty.

If you want to compute the accuracy, you have to take the \hat{y} values and convert them to â€śreal or fakeâ€ť predictions and that will give you actual accuracy numbers.

realPred = (\hat{y} > 0.5)

1 Like