sabareeshiyer / transfer-learning-using-vgg16-in-keras Goto Github PK

What made you decide not to use adam optimizer?

My understanding is that when you are done with the training and validation, you need to evaluate the model
based on test data which is a set of isolated images. This will determine whether or not the training has overfitted.

On first notebook, there were several augmentations.

But on the second notebook, there was just one augmentation.

This makes not equal generalization.

Was there a reason for this?

I used your same code for creating bottleneck. I have the following samples.
nb_train_samples = 9920
nb_validation_samples = 1376

train_top_model() after running this code got some error as below:
ValueError: Input arrays should have the same number of samples as target arrays. Found 9920 input samples and 2000 target samples.

Why does this happen?

can you explain optimizer=optimizers.SGD(lr=1e-4, momentum=0.9)?