Trying this ques,for quite some time now and cant get the desired accuracy.
I guess, there is some issues with my hyper parameters of the AlexNet Code.
Please have a look at it and guide me where I am doing wrong
https://colab.research.google.com/drive/111ROij5_im8IjaM2GSB1wHO5HcBCzu_S
Hey Anubhav,
12% is the threshold accuracy for the Pokemon Classifier Challenge. To have a better accuracy, you should use Transfer Learning techniques. They really help with that. Your hyper parameters look fine here.
I have used MobileNet Transfer Learning too. But can only improve the accuracy by another 1 or 2%.
@rachitbansal2500
Oh. Okay. I will try and look closely into the hyper parameters you have used and will get back to you shortly.
Until then please move ahead with the course.
Anubhav plz normalize data then feed the network (x = x/255 for one pixel). There will be a drastic change in output.
also keep the practice of ploating accuaracy and eror plot after training to figure out overfitting
because u trained ~1200 epochs which are likely to overfit
Okay, let me normalize the data and then see. I will also plot the accuracy and error this time to check for overfitting.
Thank You
