Cloud Machine Learning – Google Developer’s Tutorial for TensorFlow Serving


In this lab, you will use Google Cloud Platform (GCP) for training and hyperparameter tuning, TensorFlow for building a neural network model, and TensorFlow Serving for serving the trained model. This way you can get your trained model from research into production quickly and easily.

This lab is based on the first chapter of the “Machine Learning with TensorFlow on Google Cloud Platform” specialisation on Coursera. So if you want to know more about the theory behind it, make sure to check that out!

In this tutorial, we will learn how to train a neural network using Google Cloud ML Engine and make predictions using a REST API. In order to do so, we will first create a computer instance on GCP that comes preinstalled with the necessary software libraries needed for this tutorial. Then we will use TensorFlow to build a feed forward neural network model that learns from the MNIST dataset and make predictions on new images of handwritten numbers. After that we will test our model locally by sending requests via gRPC/REST to TensorFlow Serving and display some predictions in our notebook. Finally, we will deploy our trained model to Cloud ML Engine where it can be used by other applications via a REST API.

Machine Learning (ML) is one of the fastest growing areas of computer science, with far-reaching applications. The aim of this tutorial is to make you aware of the practical aspects of Machine Learning, and how to apply it in various situations.

The tutorial will guide through the basics of machine learning, but concentrate on solving a problem with Google Cloud Platform.

You’ll learn about:

How to approach a machine learning problem

How to use Google Cloud Platform for machine learning

How to implement k-means clustering on Google App Engine

How to implement a simple neural network using Google TensorFlow

This tutorial is an excerpt from “Machine Learning on the Google Cloud Platform” by Markus Noga & Lauren Murray (O’Reilly). Copyright 2018 Google, Inc., 978-1-491-98181-0.

Introduction

This tutorial shows you how to use TensorFlow Serving components running in Docker containers to serve the TensorFlow ResNet model and how to deploy the serving cluster with Kubernetes. The tutorial uses the MNIST dataset and a Jupyter notebook.

We’ll cover the following:

Brief overview of the ResNet model and MNIST dataset

Prepare the training environment with Jupyter notebooks running on Google Cloud Platform (GCP)

Train and export a simple linear model using TensorFlow Serving signatureDefs API

Build Docker images, push them to GCR, and run them locally in Docker containers

Set up a Kubernetes cluster on GKE and deploy your trained model to it

This year, Google is hosting its annual Google Cloud Platform Online Coding Competition from March 28th to May 1st. This competition will test your knowledge of the Google Cloud and will give you the chance to win awesome prizes.

You and up to four teammates will build a project in the cloud that solves a problem or improves a process for one of Google’s non-profit partners. The project should leverage at least one of the following: App Engine, Compute Engine, Container Engine, Cloud Datastore, Cloud SQL, BigQuery, or Cloud Storage. You’ll also need to use at least one other Google product such as Maps, YouTube, or Calendar.

After submissions close on May 1st, an esteemed panel judges will review all submissions and choose the top 10 finalists. Finalists will be announced on May 16th and winners will be announced on June 15th.

There are over $100K in prizes available for the competition including $10K in cloud credits and 4 tickets to Google I/O 2017!

Machine Learning (ML) has been a hot topic in the technology industry for a while now and has gained significant traction in the last few years. One of the most promising areas of ML is Deep Learning (DL). Deep learning has shown its power in several application areas including image recognition, text analytics, and information retrieval.

TensorFlow is an open source software library for high performance numerical computation. Created and released by Google in 2015, it has seen tremendous growth and popularity in the Deep Learning community. TensorFlow is a symbolic math library based on dataflow and differentiable programming. It uses multi-dimensional arrays to represent the data. Multidimensional arrays are also called tensors and that’s where TensorFlow gets its name from!

This tutorial will show you how to use your TensorFlow models in a browser using JavaScript, as well as helping you get started with training your own models on Cloud Machine Learning Engine (CMLE). The first step will be to build a model using TensorFlow’s Python API, then we’ll look at serving that model from Python and from Node.js via gRPC. Finally, we’ll train a model using CMLE so you can see how easy it is to scale your training jobs to handle massive amounts of


Leave a Reply

Your email address will not be published.