Loss function value

Hi ,
I just wanted to know that what is better, loss function with higher value of loss function with lower value for training and test dataset??

Hi @Shreyash_Gupta , sorry but I am guessing your question is not clear to me. In the training process, we are ever trying to minimize the loss function and so, when you say “loss function with higher value”, I really don’t understand your point. Maybe if you could elaborate :slightly_smiling_face:

Hi , thanks for replying,

When I am fitting the model with train images and labels the value of loss function is reducing by every Epoch.

But when I use test image and label to test the model the value of loss function is higher than the loss function in training model.
Why is that?

And What is better loss function with high value or loss funciton with lower value?

Ok, now I got your point. It seems you have some doubts about bias, variance, overfitting, and underfitting. Basically, a good model will have a low bias and low variance. But because the explanation is a bit longer and checking same examples is a good way to understand the scenario, I suggest you take a look the Andre Ng’s free book named Machine Learning Yearning which you can download from this link https://www.deeplearning.ai/programs/. Go to chapters 20, 21, 22, see the examples and you will understand how to interpret the behavior of your training model.
Keep learning!

1 Like

Thank you for this resource. It’s very helpful.

I hope that helped you with your query

1 Like