Clarify that the model needs to be trained to be meaningful

[xuhdev]

The Docker image is immediately deployable. But the document does not make it clear that user should have trained the model first, and otherwise the immediately deployable model isn't really useful. This can be confusing for some users who are not familar with recommender systems.