You can see in the previous command we mentioned something called bottleneck, so what is a bottleneck?

So the first step that will be taken in order to get our retrained model is to analyse all the images that we have provided and calculate the bottleneck values for each one of them.

As we mentioned that we are using Inception V3 as a pre-trained model. Inception V3 has many layers that are lying on top of each other. These layers are already pre-trained and have information that can be used to discriminate between the images. We are only training the last fully-connected layer (the one just before the output layer) and all the other previous layers will retain their trained state.

So what's a bottleneck? It is a term that TensorFlow uses to refer to the layer just before the final layer that is actually responsible for the classification. So any image in the training set is used multiple times during the training and calculating the layers behind the bottleneck takes large time for each image. So by caching the outputs of these lower layers on the disk we are avoiding wasting a significant amount of time. By default, the bottleneck is stored in /tmp/bottleneck directory.