Module 4: Deploy

Module Overview

This module focuses on deploying neural networks to production environments. You'll learn how to save and export trained models, convert them to optimized formats for deployment, and serve them using various deployment options. The module covers strategies for model monitoring, maintenance, and scaling to handle real-world workloads. By the end of this module, you'll be able to take your trained neural networks and make them accessible for real-world applications.

Learning Objectives

1. Understand model export formats and optimization

  • Compare different model saving formats (HDF5, SavedModel, TFLite)
  • Optimize models for deployment using quantization
  • Convert models to TensorFlow Lite for mobile and edge devices
  • Apply pruning to reduce model size without sacrificing accuracy

2. Implement deployment strategies

  • Deploy models using TensorFlow Serving
  • Create REST APIs for model inference
  • Containerize models using Docker for portable deployments
  • Implement batch and real-time inference pipelines

3. Monitor and maintain deployed models

  • Implement systems to track model performance in production
  • Detect and respond to model drift
  • Design strategies for model updates and version control
  • Create A/B testing frameworks for model comparison

Guided Project

Deploying Neural Networks

GitHub Repo Slides

Guided Project File:

DS_424_Deploy_Lecture.ipynb

Module Assignment

Please read the assignment file in the GitHub repository for detailed instructions on completing your assignment tasks.

Assignment File:

DS_424_Deploy_Assignment.ipynb

In this assignment, you will take the sketch classification model you've been developing and deploy it for real-world use. Your tasks include:

  • Saving and loading models in different formats
  • Optimizing models for deployment using quantization
  • Converting models to TensorFlow Lite
  • Creating a simple REST API for model inference
  • Implementing monitoring for deployed models
  • Building a simple user interface for interacting with the model

Assignment Solution Video

Check for Understanding

Complete the following items to test your understanding:

  • Compare and contrast different model saving formats and their use cases
  • Explain the benefits and trade-offs of model quantization
  • Describe the process of creating a REST API for model inference
  • Outline strategies for monitoring model performance in production
  • Explain how to implement A/B testing for model comparison
  • Describe approaches to handling model drift in production environments

Additional Resources

Model Export and Optimization

TensorFlow: Saved Model Guide TensorFlow: Post-Training Quantization TensorFlow: Pruning Models

Deployment and Production

TensorFlow Extended (TFX) Guide TensorFlow Serving Documentation Flask Documentation