So, now that I've got you excited about all the possibilities of using machine learning, let's actually get into the specifics. We are going to start by looking at the world of ML model types, and why some out of the box models are perfect starting points. Now, there are two broad categories of ML models. You have models that are already built,and called pre-trained, and then those custom models that you build yourself using code like tensorflow. For pre-trained models, google has already figured out a lot of these hard problems, things like, the vision API that you saw before or an images, the speech API which is trained on something like, Youtube captions, and the machine translation API for things like language translation. They recall that how well your model is trained depends largely on how much data you have access to. As you might expect, Google has a lot of images and text and e-mail researchers to train its pre-built models, so you can use those instead of reinventing the wheel. As an example, if you're looking to have video captions included in a recent webinar that your company hosted, consider using the translator speech APIs, instead of trying to hand code and build the language recognition and model for yourself. Another example, if you have text documents, log expense receipts, that need to be classified by expense type, consider using something that the Cloud Vision API for OCR, so you can then mind the text from the receipts, then drop that data into something like Big Querrey for analysis. In your next lab, you'll actually be invoking these APIs through Cloud data lab, and I bet you'd be surprised on how easy it is to call those ML models. Now, all that said, if you do want to go to the custom model route and write some of your own tensorflow code, which is taught in a lot of other future data engineering courses via on this one, note that Cloud machine learning engine as you see all the way there on the right, will fully manage, and host, and run your custom build models. So even if you want to write your own code, the Cloud can still be a place where you can store and write. Here's a key point, the ML APIs will be covering throughout this course like cloud vision, the abstract way a lot of the underlying model code, and leave you with a simple service that you can provide in that input and receive label output. The bottom line here is that, even when we eventually build our own models, interacting with them for things like, predictions should be simple, scalable, and seamless. You ultimately want to provide the end user, or whoever is invoking your service or your model through like an API, a simple interaction interface like, give me input to find the expected parameters and then generate output that meets our performance qualifications.