Inspecting ML models using heat map and TCAV.
In the HW3_Question1_FINAL is where I built a custom CNN for recognizing hand-written digits from cifar10 dataset as well as another custom CNN that has Resnet50 for its base (Transfer Learning).
These 2 models are trained and the models are saved (to be used in the other file) as well as their fitting histories (to be used in plotting if a session expires)
In the HW3_Question2_FINAL, I created heatmap for my CNN model from the HW3_Question1_FINAL file on a random image of a horse.
Finally, the most exciting part TCAV did not work. I tried to run it on cifar100 images of cloud. However, it seems that TCAV is not capable to cooperate with tensorflow 2 and tf.keras models...
I did not follow the main repo that they propose since it is fitted for ImageNet. This post has its own version. However, this could not work with tensorflow 2... Did not know how to fix it...