Reviews of Codeless Deep Learning with KNIME

However, the fact that these libraries and frameworks are code-driven, meaning that in order to use them to leverage deep learning in your project you must already have a solid understanding of how to program, renders them inaccessible to a large group of potential users; specifically, those who do not code. But what if you intend to understand the variety of deep learning algorithms and architectures and how to use them, but have no intention of learning to code?

In the Venn diagram of those who could take advantage of deep learning and those who know how to code, the book Codeless Deep Learning with KNIME by Kathrin Melcher and Rosaria Silipo is specifically aimed at the section of the "those who could take advantage of deep learning" circle not intersecting "those who know how to code." It may seem like a tall order - understanding and implementing deep learning solutions without coding - but when you realize that you can focus on only having to learn one of these two non-trivial time-consuming tasks, one could imagine that it may actually be advantageous.

Now, about the book: Codeless Deep Learning with KNIME starts out by introducing KNIME Analytics Platform, software for implementing, among other analytics solutions, neural networks. The difference between this implementation software and the previously-mentioned libraries and frameworks is that it is equipped with an intuitive graphical interface, making it workable for non-programmers. An overview of the platform is provided, with the basics being covered.

The book then moves on to neural networks themselves, explaining the concepts while demonstrating how to use KNIME to implement them. It progresses from the very basics, to more complex feed forward networks, to autoencoders and recurrent networks and beyond, all the while explaining how these different networks can be used in the real world, with examples from image classification to fraud detection to demand prediction and more. The deployment of the resulting models is also given treatment, something often overlooked completely in such a book.

Advanced concepts are also treated holistically. Topics such as natural language processing are really collections of disparate concepts that can be brought together to perform lots of related tasks, which can, in total, initially be difficult to grasp. Codeless Deep Learning with KNIME introduces these ideas in full start-to-finish tutorial style, so you can build end-to-end solutions within the same chapter, putting it all together - and doing so without code. Two other well-known use cases for deep learning - Image classification with convolutional neural networks and machine translation - are also treated in this comprehensive style.

Given that you don't need to learn to code to take advantage of deep learning, thanks to KNIME's platform and Melcher and Silipo, you could really go from zero to hero with Codeless Deep Learning with KNIME rather quickly. As someone who has learned to implement neural networks the old-fashion way, through the trial and error of programming, I can attest that eliminating that aspect - as well as the need to learn to program altogether! - will make learning and employing these concepts quicker and easier.

If you are looking to take advantage of deep learning without first learning how to code, Codeless Deep Learning with KNIME is for you.

In the first section, chapter one provides all you need to know to get started with KNIME and its deep learning extension and provides a quick introduction to KNIME itself. Note that if you’re a complete novice to KNIME, this isn’t sufficient to bring you up to speed for the level of proficiency required to tackle the rest of the book. In that case, you should first develop a minimal experience in KNIME using tutorials and examples. Chapter two provides many useful examples on data manipulation to get you started in the right direction.

A codeless approach to deep learning is very appealing, but it doesn’t mean that you don’t require understanding of the underlying maths and algorithms. Luckily, chapter three provides a brilliant and extremely clear summary of the fundamental theory of Artificial Neural Networks (feedforward propagation, backpropagation, gradient descent, etc.) and with a compendium of the important hyper-parameters for the most popular optimizers. This chapter is worth reading, in general, if you wish to understand deep learning regardless of which tool or language you choose for implementing your solutions.

Chapter four is the first step toward more complex deep learning architectures with a guided tutorial to build and train a general feedforward neural network. You must master this aspect before you can proceed with the more modern and specialized deep learning architectures.

Section two (chapters five to nine) deals with the most important and modern architectures for deep learning. Each chapter comes with a fully developed example of a real-world application: Autoencoders, Recurrent Neural Networks, Long Short-Term Memory, Convolutional Neural Networks, with applications in fraud detection, demand prediction, natural language processing, machine translation, and image classification.

Finally, section three (chapters ten and eleven) provides the explanation and the guidelines for the deployment of a solution using KNIME Server.

This book isn’t for the fainthearted and occasional data wrangler - it’s for the fearless KNIME data miners who prefer the elegant and expressive visual workflows of KNIME over the messy stream of consciousness of Python notebooks.

I liked the way the topics are laid out and sequenced: starting with the basics and foundations of KNIME, and then diving into deep learning with the extension, covering different deep learning architectures, and their codeless implementation in KNIME Analytics Platform.

The coverage is comprehensive, motivational, conceptual, and with just about enough algorithmic/mathematical explanations and details.

In summary, this book is one of a kind in that it explains highly technical, mostly code-based, deep learning algorithms in an easy to understand manner, using a visual modeling paradigm based on KNIME Analytics Platform.

The authors provide the right mix of theoretical foundation followed by practical case studies on fraud detection, natural language processing (NLP), image classification, and more. The last section (two dedicated chapters) covers the usually neglected topics, such as implementing trained models to production and a much-needed chapter on deployment best practices. The deployment options include a web app or a REST web-service - again without any coding and completed in no time. However, it requires access to a licensed version of KNIME Server.

To summarize, if you learn best by doing, you can't go wrong with this book. I recommend this to someone who wants to get started but may feel a little lost and anxious. You'll want to check out the book's companion workflows on the KNIME Hub. It contains lots of workflows with configured nodes and data that you can run right on your KNIME Analytics Platform.

One of the most important things in any data science project is cleaning up the data - it's often the most time-consuming part, and this book clearly guides you through that process. The sample project that generates new fairy tales was interesting, but given the output I'm not too worried about computers taking over all the authoring jobs any time soon.

The first section, which introduces the reader to KNIME and the theme of neural networks, allows the reader to enter the fantastic world of KNIME and Deep Learning without ever having had to deal with it before. In my case, the first section was a pleasure reading. For me it was used to dust off the deep learning theory. Moreover, it starts to introduce an important aspect - one which many data scientists or operators in the field often forget is the main part: data preparation. In my opinion the last section, deployment and productionizing, is very well done. I recommend to anyone working in the field of data analysis, artificial intelligence, deep learning, etc. to read it.

The second section was challenging but very interesting reading - discovering different methodologies and fields of application such as the Convolutional Neural Network for image classification, which was fantastic. I also really liked the part dedicated to NLP and Translation. And even if I knew the topic well, it was very useful to see the KNIME application.

In conclusion, I highly recommend the book to all those who are new to deep learning and those who already know it but want to see how to apply it without using a line of code. A very nice and enjoyable book, something that it is difficult to find in our day on the topic.

The book walks the reader through KNIME and its key processes with many examples and content related to deep learning theory. After the reader learns about the basics of the KNIME framework and the theory of deep learning, several sections provide practical examples with the aim of putting everything together.

Sections of the practical applications include natural language processing, image analysis, machine translation, fraud detection and energy demand prediction. For each application the authors provide information on the necessary steps of pre-processing, analyzing and testing. Finally, a special section is given to the deployment phase of a project, which is often not well explained in many data science books.

The pros:

• Well-written book, explains key concepts and gives best practices
• Many practical applications provided
• Extensive information on Deployment options.

The cons:

• I’d prefer an example of natural language processing such as entity detection (a technique that transforms unstructured data to a structured format) to be given instead of the free text generation example
• Although the KNIME framework has so many of them, it would be nice if the authors simply mentioned - rather than fully explain due to limited space - more KNIME nodes that are likely to be of use when it comes to accessing and pre-processing data. This addition could prove very useful since with programming a lot of pre-processing actions can be implemented and programming is not part of this book.

Overall, a very useful and well-written book.