Architecting Production-ready ML Models Using Google Cloud ML Engine

This course covers Cloud ML Engine, a powerful service that supports distributed training and evaluation for models built in TensorFlow, Scikit-learn and XGBoost.
Course info
Level
Intermediate
Updated
Jan 9, 2019
Duration
2h 12m
Table of contents
Course Overview
Introducing Google Cloud ML Engine
Deploying XGBoost Models to Cloud ML Engine
Deploying Scikit-learn Models to Cloud ML Engine
Deploying TensorFlow Models to Cloud ML Engine
Description
Course info
Level
Intermediate
Updated
Jan 9, 2019
Duration
2h 12m
Description

Building machine learning models using Python and a machine learning framework is the first step towards building an enterprise-grade ML architecture, but two key challenges remain: training the model with enough computing firepower to get the best possible model and then making that model available to users who are not data scientists or even Python users. In this course, Architecting Production-ready ML Models Using Google Cloud ML Engine, you will gain the ability to perform on-cloud distributed training and hyperparameter tuning, as well as learn to make your ML models available for use in prediction via simple HTTP requests. First, you will learn to use the ML Engine for models built in XGBoost. XGBoost is an ML framework that utilizes a technique known as Ensemble Learning to construct a single, strong model by combining several weak learners, as they are known. Next, you will discover how easy it is to port serialized models from on-premise to the GCP. You will build a simple model in scikit-learn, which is the most popular classic ML framework, and then serialized that model and port it over for use on the GCP using ML Engine. Finally, you will explore how to tap the full power of distributed training, hyperparameter tuning, and prediction in TensorFlow, which is one of the most popular libraries for deep learning applications. You will see how a JSON environment variable called TF_CONFIG is used to share state information and optimize the training and hyperparameter tuning process. When you’re finished with this course, you will have the skills and knowledge of the Google Cloud ML Engine needed to get the full benefits of distributed training and make both batch and online prediction available to your client apps via simple HTTP requests.

About the author
About the author

An engineer and tinkerer, Vitthal has worked at Google, Credit Suisse, and Flipkart and studied at Stanford and INSEAD. He has worn many hats, each of which has involved writing code and building models. He is passionately devoted to his hobby of laughing at his own jokes.

More from the author
More courses by Vitthal Srinivasan
Section Introduction Transcripts
Section Introduction Transcripts

Course Overview
Hi, my name is Vitthal Srinivasan, and I'd like to welcome you to this course on architecting Production-ready ML Models Using the Google Cloud ML Engine. A little bit about myself, I have master's degrees in Financial Math and Electrical Engineering from Stanford University and have previously worked in companies such as Google in Singapore and Credit Suisse in New York. I am now co-founder at Loonycorn, a studio for high-quality video content based in Bangalore, India. Building machine learning models using Python and an ML framework is the first step towards building an enterprise-grade ML architecture. But two key challenges still remain, training the model with enough compute firepower to get the best possible model and then making that model available to users who are not data scientists and may not even be Python users. You will gain the ability to perform on- cloud distributed training and hyperparameter tuning, as well as to make your ML models available for use in prediction via simple HTTP requests. First, you will learn to use the ML engine for models built in XGBoost. XGBoost is an ML framework that utilizes a technique known as ensemble learning to construct a simple, strong model by combining several weak learners as the unknown. XGBoost is very popular in the Kaggle community, and it makes a lot of sense for use on the cloud because it features a fast and reliable training algorithm that converges quickly. On the cloud, time is money, and so the speed of training is important. Next, you will discover how easy it is to port serialized models from on premise to the GCP. You will build a simple model in scikit-learn, which is the most popular classic ML framework, and then serialize that model and port it over for use on the GCP using ML Engine. Finally, you will explore how to tap the full power of distributed training, hyperparameter tuning, and prediction in TensorFlow, which is the most popular library for deep learning applications right now. TensorFlow was the first library that ML Engine supported, and both originated at Google, so the integration between these two technologies is very tight. You will see how a JSON environment variable called TF_CONFIG is used to share state information and optimize the training and hyperparameter tuning processes. Once you're finished with this course, you will have the skills and knowledge of the Google Cloud ML Engine needed to get the full benefits of distributed training and make both batch and online prediction available to your client apps via simple HTTP requests.