Today is going to be tough! I'll be running directly from work to rehearsal at the Palace of Fine Arts for a City Dance benefit performance. Also, right around the time we go on stage tonight the city is going to explode from Game 4 of the NBA finals. I'm writing this from an Uber right now.

Now that the Kaggle data is sorted, training and validating our model is as simple as running the seven lines of code we've seen before:

from vgg16 import Vgg16
batch_size = 64
vgg = Vgg16()
batch_training = vgg.get_batches("/data/train/", batch_size=batch_size)
batch_valid = vgg.get_batches("/data/valid/", batch_size=batch_size)
vgg.finetune(batch_training)
vgg.fit(batch_training, batch_valid, nb_epoch=1)

Hmm. You might have noticed that I only mentioned training and validating our model, and that the code itself only contains variables for batch_training and batch_valid. How are we supposed to pass the model our test data?

Because the functions helpfully provided by the course creators have been doing most of the work up until this point, I am actually pretty clueless on how a folder full of images gets passed to a deep learning model.

vgg.fit won't work, the test data isn't sorted into categories (and it wouldn't be correct to fit the model on our training data anyway).

vgg.predict doesn't work, telling me it expects an array.

In the course forums, I find advice from a past student to "read the dogs-cats-redux notebook".

Found it!

Oh shit.