You're reading from Codeless Deep Learning with KNIME Build, train, and deploy various deep neural network architectures using KNIME Analytics Platform

Product type Paperback

Published in Nov 2020

Publisher Packt

ISBN-13 9781800566613

Length 384 pages

Edition 1st Edition

Languages

Python

Tools

Knime

Concepts

Deep Learning

Authors (3):

Kathrin Melcher

KNIME AG

Rosaria Silipo

View More author details

Table of Contents (16) Chapters

Preface

1. Section 1: Feedforward Neural Networks and KNIME Deep Learning Extension

2. Chapter 1: Introduction to Deep Learning with KNIME Analytics Platform FREE CHAPTER

3. Chapter 2: Data Access and Preprocessing with KNIME Analytics Platform

4. Chapter 3: Getting Started with Neural Networks

5. Chapter 4: Building and Training a Feedforward Neural Network

6. Section 2: Deep Learning Networks

7. Chapter 5: Autoencoder for Fraud Detection

8. Chapter 6: Recurrent Neural Networks for Demand Prediction

9. Chapter 7: Implementing NLP Applications

10. Chapter 8: Neural Machine Translation

11. Chapter 9: Convolutional Neural Networks for Image Classification

12. Section 3: Deployment and Productionizing

13. Chapter 10: Deploying a Deep Learning Network

14. Chapter 11: Best Practices and Other Deployment Options

15. Other Books You May Enjoy

Leave a review - let other readers know what you think

Conversion of the Network Structure

The goal of a deployment workflow is to apply a trained network to new real-world data. Therefore, the last step of the training workflow must be to save the trained network.

Saving a Trained Network

All networks described in this book have been trained using the Keras libraries, relying on TensorFlow as the backend. So, the most natural way to save a network is to continue using the Keras libraries and therefore to use the Keras Network Writer node. The Keras Network Writer node writes the network, including its weights, in Keras format into a .h5 file.

However, Keras-formatted networks can only be interpreted and executed via the Keras libraries. This is already one level on top of the TensorFlow libraries. Executing the network application on the TensorFlow Java API directly, rather than on a Python kernel via the Keras Python API, makes execution faster. The good news is that KNIME Analytics Platform also has nodes for TensorFlow execution...