Course CPB102:
Google Machine Learning with Cloud ML

(1 day)


Course Description

Through a combination of instructor-led presentations, demonstrations, and hands-on labs, students learn machine learning and Tensorflow concepts and develop hands-on skills in developing, evaluating, and productionizing machine learning models.

This 1 day instructor-led course builds upon CPB100 and CPB101 (which are prerequisites).


  • Knowledge of Google Cloud Platform Big Data & Machine Learning Fundamentals to the level of CPB100
  • Knowledge of BigQuery and Dataflow to the level of CPB101
  • Knowledge of Python and familiarity with the numpy package
  • Knowledge of undergraduate-level statistics to the level of Udacity ST101

Learning Objectives

  • Understand what kinds of problems machine learning can address
  • Build a machine learning model using TensorFlow
  • Build scalable, deployable ML models using Cloud ML
  • Know the importance of preprocessing and combining features
  • Incorporate advanced ML concepts into their models
  • Employ ML APIs
  • Productionize trained ML model

Who Should Attend

This class is intended for programmers and data scientists responsible for developing predictive analytics using machine learning. The typical audience member has experience analyzing and visualizing big data, implementing cloud-based big data solutions, and transforming/processing datasets.


Module 0: Welcome [⅓ hr]

We assume that attendees may attended CPB100.

  • Logistics
  • Introductions

Module 1: Getting Started with Machine Learning [1½ hr]

  • What is machine learning (ML)?
  • Effective ML: concepts, types
  • Evaluating ML
  • ML datasets: generalization
Lab: Explore and create ML datasets

Module 2: Building ML models with Tensorflow [2 hr]

  • Getting started with TensorFlow
Lab: Using tf.learn
  • TensorFlow graphs and loops + lab
Lab: Using low-level TensorFlow + early stopping
  • Monitoring ML training
Lab: Charts and graphs of TensorFlow training

Module 3: Scaling ML Models with CloudML [1 hr]

  • Why Cloud ML?
  • Packaging up a TensorFlow model
  • End-to-end training
Lab: Run a ML model locally and on cloud

Module 4: Feature Engineering [1.5 hr]

  • Creating good features
  • Transforming inputs
  • Synthetic features
  • Preprocessing with Cloud ML
Lab: Feature engineering

Module 5: ML Architectures [optional]

  • Wide and deep
  • Image analysis
  • Embeddings and sequences
  • Recommendation systems
  • Summary