I got through the assignment easily enough, but at the end of the assignment, the authors say “This model is obviously overfitting the data”. I would like to understand what exactly it means to overfit, and what makes it obvious in the case of this model?
The model is overfitting as the training accuracy is close to 100% and the test accuracy is 70%, this implies that the model is very good when trying to recognize the images used during the training / learning phase, i.e. images that the model has “seen” already but it performs much worse with new / unseen images.
For an example of misclassification you could try with index = 5
Ahh, got it. I had thought it would be something like that, but it hadn’t struck me that the difference between test and training accuracy will indicate overfitting. Thanks a lot for your help!
on week 2 assignment i am getting assertion error can u help me
Hi @charantej, as your issue has nothing to do with this particular question I think you should create a new post with your question and the detailed error you are getting explaining in which exercise you are having issues and including the error log.
We have 12288 + 1 fitting parameter and 209 training samples. This means there are many solutions that make the costfunction exactly 0 (minimal). So overfitting looks very natural to me, and must actively be avoided. How do we do that? By choosing bad initial values, stopping arbitrarely the iteration? Choosing a ‘bad’ learning parameter?
Will this be adressed in later classes, courses?