Keynote Google - Machine Learning APIs for Python Developers

published May 19, 2017

Keynote talk from Google about Machine Learning APIs for Python Developers, at PyGrunn.

See the PyGrunn website for more info about this one-day Python conference in Groningen, The Netherlands.

Lee Boonstra and Dmitriy Novakovskiy give this talk, they work for Google Cloud, one of the gold sponsors of PyGrunn.

Python at Google

Google loves Python. :) It is widely used internally and externally. We are sponsoring conferences. We have open source libraries, like a Google Data Python Client Library, libraries for youtube, app engine, etcetera. We use it for build systems, report generation, log analysis, etc.

How can you use Google Cloud Platform for your app or website? You can deploy at scale. You can embed intelligence empowered by machine learning, we provide multiple pre trained models. You can use serverless data processing and analytics.

Machine learning

Let me explain it in a simple way. You want to teach something to a kid: what is a car, what is a bike? You point at a car or bike and explain what it is called. With machines we shoot in lots of data and they start to see patterns.

  • Artificial intelligence: process of building smarter computers
  • Machine learning: process of making a computer learn

Machine learning is much easier.

Our CEO: "We no longer build mobile-first applications, but AI-first."

We have a lot of data, better models, and more computing power. That is why machine learning is happening a lot now.

Google created the open source Tensorflow Python framework for machine learning. And we have hardware to match. We have ready to use models for vision, speech, jobs, translation, natural language, video intelligence.

  • Vision API: object recognition (landmarks), sentiment on faces, extract text, detect inappropriate content. Disney game: search with your phone for a chair, and we show a dragon on the chair. Point your camera at a house, and you see a price tag.
  • Speech API: speech recognition (write out the transcript, support for 80 languages).
  • Natural language API: really understand the text that is written, recognise nouns and verbs and sentiment.
  • Translation API: realtime sub titles, automatic language translation using more context than earlier versions.
  • Beta video intelligence: label detection, enable video search (in which frame did the dog first appear).


Go to the Google Cloud console and create a free account to play with. You need to enable the APIs that you want to use. Install the command line tools if you want to run it on your local machine. And pip install google-cloud.

We use machine learning for example in GMail to show you a possible answer to send for an email you receive.

Walkthrough of machine learning and TensorFlow

Google Cloud Dataflow. Dataflow is a unified programming model for batch or stream data processing. MapReduce-like operations. Parallel workloads. It is open sourced as Apache Beam, and you can run it on Google Cloud Platform.

You put files in Cloud Storage. Process this in batches, with Python and Dataflow. This uses pre-trained machine learning models. Then store results in BigQuery, and visualize the insights in Data Studio.