README
¶
Introduction
Google's AI platform brings a number of tools together to enable you to create end-to-end machine learning pipelines. This repository provides an example of using the platform to train a recommendation engine using the Go programming language.
Motivation
At the time of writing, the AI platform is relatively new and as a result the documentation is incomplete, particularly with respect to Go. The AI platform is not part of the Google Cloud SDK. Instead, we need to use the Google APIs client which is a bit 'lower level', in that it just exposes Google's various REST APIs in raw form. Documentation for using the AI platform with Python is useful as a basis for doing the same thing in Go.
n.b. although not part of the SDK, the AI platform's functionality is still available through the gcloud command line tool, it just isn't available as part of the SDK for Python, Go, etc.
Accessing the API
The AI platform's API is described here: https://cloud.google.com/ml-engine/reference/rest. To set up the client do something like this:
import "google.golang.org/api/ml/v1"
client := /* construct oauth'd HTTP client */
service := ml.New(client)
Authentication is covered further down.
Regions
Some features are only available in certain regions; for instance TensorFlow.
REST endpoints
Using the APIs client the endpoints are built up in stages that follow the layout of the REST endpoints. For example:
service.Projects.Jobs.List("projects/<your gcloud project name>").Do()
One thing to watch out for is the project name: it must be prefixed with projects/, or else you'll get a 404 not found error like The requested URL <code>/v1/<your gcloud project name>/jobs?alt=json&prettyPrint=false</code> was not found on this server.
Authenticaton / OAuth
To make API calls you'll need to be authenticated using OAuth. There are plenty of ways to do this and some are quite complex. We cover two simple examples:
- Using the application-default context. The application-default context is the most common and simple way to authenticate against Google APIs.
- Directly using a service account by loading a key. This is provided as an example but it's not recommended, as application-default should be preferred for accessing the AI platform.
Directories