The testing and debugging guidelines in this course can be complex to implement. You can implement some of the guidelines using TensorFlow and TensorFlow Extended (TFX). TFX is an end-to-end ML pipeline based on TensorFlow. For a demo, view this end-to-end TFX example. To complement the end-to-end example, the following table lists available resources in TF and TFX by guideline. Only guidelines supported by TF or TFX are listed.
| Guideline | TF/TFX Implementation | |
|---|---|---|
| Guidelines for debugging your ML model | ||
| Exploring your data to understand it | Explore your data using Pandas or Facets.
|
|
| Validating input data using a data schema | Use TensorFlow Data Validation. | |
| Implementing tests for ML code | First, debug your TF models with Eager Execution. Then write tests with Tensorflow Testing. | |
| Metrics | ||
| Generating model metrics | TensorBoard visualizes your TF graph and plots metrics. See Tensorboard: Graph Visualization. | |
| Deployment to Pipeline | ||
| Testing model quality in production | Use Tensorflow Model Analysis. | |
| Checking for training-serving skew | Avoid feature skew by sharing feature engineering code across training and serving by using TFX Transform. | |
| Tracking model staleness | -- | |