Google Developer Days (GDD) is a global event that showcases Google’s latest Developer products and platforms. It’s designed to help you quickly develop great apps, build and retain an active user base, and make the most of your tools to make more money. The 2018 Google Developers Conference was held in Shanghai on September 20 and 21. Google Developer Conference 2018 Gold Nuggets at 👉
This post is a review of the presentation “Deploying Machine Learning in a production Environment with TensorFlow Extended (TFX)” delivered by Tinhao Qiu, a Software engineer at Google Brain, on September 21, 2018.
TensorFlow Extended
The structure of the TensorFlow Extended platform is shown in the figure above. The following sections are based on the four parts of Google open Source.
Data Conversion (TensorFlow Trainsform)
Before using a data set, it is often necessary to transform the data, and TFT allows you to transform the data into TF diagrams, as well as define special transformations to express special features. The TF calculation graph generated is sealed and contains all eigenvalue information, which means it can be applied to training and deployment. This is a very important point. In the production environment, we may use different languages and paths for training and deployment, but the TF calculation chart is sealed, which solves this problem well.
Example:
Taxi tipping example
GitHub:github.com/tensorflow/…
Model training (Estimators or Keras)
TensorFlow Estimators were used for model training
The trained model will be output as SavedModel.
Tf.keras was used for model training
TensorFlow Model Analysis
After model training, it is often necessary to rewrite the code to evaluate and validate the model. TFMA emerged to solve this problem.
After running the code shown above, a graphical interface like the one shown below will appear, making it more intuitive to evaluate the model.
GitHub:github.com/tensorflow/…
Model Deployment (TensorFlow Serving)
After deploying the TensorFlow Serving model, you can access it using the Serving interface.
GitHub:github.com/tensorflow/…
TensorFlow Data Validation
The Data Validation module consists of the following:
Calculate statistics and visualize statistics
Derive the initial Schema
Data validation
Training/online data deviation detection
GitHub:github.com/tensorflow/…
conclusion
The process for using TFX in a production environment is as follows:
TFX: A Tensorflow-based Production-Scale Machine Learning Platform, KDD (2017)
That is all the content of this speech, I hope it will be helpful to you. 