gen = generator()
loss, mae = model.evaluate_generator(gen, 1000)
#This produces Test set MAE:  1.80

gen = generator()
predictions = model.predict_generator(gen, 1000).flatten()

gen = generator()
y_test = np.zeros((1000))
i = 0
while i < 1000:
    p = next(gen)[1]
    y_test[i] = p
    i += 1

# Manually compute MAE
np.mean(np.abs(predictions - y_test))

# This produces 13.74