search

whaiao / Genetic-Algorithms-on-CNNs

Project for WT2020/2021 @ ICL, Heidelberg University
1 stars 1 forks source link

Usage of code #2

Closed kadiraltug closed 2 years ago

kadiraltug commented 2 years ago

Again me sir, How can i use your code with my own data? it is only about ga.py calc_fitness function? How can i use your code on my code? Thanks.

whaiao commented 2 years ago

You will need to preprocess your data to a size of 32x32 + color channel since the code was written for CIFAR-10.

kadiraltug commented 2 years ago

I use datasets.ImageFolder instead of datasets.CIFAR10 . It is working but accuracy is so small and Loss is so high. Is there any problem or should I increase epoch size?

whaiao commented 2 years ago

The highest accuracy I got was 62.5, so it might be the case that, depending on your dataset it won't even achieve that much. You can definitely tune hyperparameters, but it was more an experiment than a really groundbreaking solution.

kadiraltug commented 2 years ago

Okey, Thanks sir

whaiao commented 2 years ago

Alright, happy tuning! If anything occurs, feel free to open another issue.