Machine Learning Automation with TPOT: Build, validate, and deploy fully automated machine learning models with Python

The author did an extensive work in demystifying ML from the basics to the ground up. Additionally, the auto ML tools in the Python ecosystem were discussed extensively as creating an ML model is not the end of the ML pipeline.More importantly, finding the best model suitable for your problem could be quite challenging. To solve this problem, the author provided deep insights about how TPOT (a python library) helps to solve this problem. Moreover, excellent tips were provided for model deployment with helpful code snippets to help with code development. If you are an data scientist or an ML professional or a researcher with a focus in AI looking to advance your skillset in ML and learn more, then I highly recommend this book for you.

This book dives deep into one of the most widely used AutoML package - TPOT with such deft and ease that any reader (novice/expert) will be pleasantly surprised. The book also covers every possible use case of the package starting from simple hands on with public datasets leading up to complex production deployment.Highlights:1) The author talks in a simple and relatable language, which makes the reader feel at ease. This is a very crucial aspect of any book aimed at explaining tools/packages , since an good understanding in the first few chapters will prevent confusion in the later ones2) I was genuinely surprised by the efforts put into explaining package installation/setup. The clarity was exceptional from simple python setup leading up to complex production scenario involving parallelization3) A special mention must be made to Chapter 5 - Parallel Training with TPOT and Dask. This chapter covers only the crucial aspects of parallelization and marries it to TPOT pipeline building. It is truly a work of art !!4) Chapter 8 is yet another excellent work. Model deployment is often not covered in many books in the AutoML field due to its perceived lack of applicability. Hence it was quite enlivening to see the author take the challenge head on and give a clear and concise pictureShortcomings:1) It is widely known that there are tens of open-source packages for AutoML and pipeline building. While the book does an excellent job in explaining TPOT, it performs poorly in highlighting its advantages over other packages. For instance, exporting the pipeline is a powerful feature in TPOT, which the author has merely touched upon without going into greater detail.2) In Chapter 6 , while the author tries to cover Neural Networks (Multi Layer Perceptrons) , there are multiple occurrences where the diagram is rather too simplistic and doesn't relate to the text content explaining it3) In the same chapter, the author also makes the mistake of unnecessarily diving into intricate details of forward pass, while completely ignoring back propagation. Strictly speaking neural nets is beyond the scope of this book, but any attempt to do so needs to be balanced, lest risk the reader being misleadThis book is a great source of information not just to learn TPOT, but to learn about AutoML pipelines (creation, tuning, deployment, maintenance) in general. The wide spectrum of use cases ranging from beginners' ride to production deployment covered in this book can be a great learning for readers

The author walks the reader through the process of discovering and optimizing machine learning pipelines, starting with a discussion of the methodologies utilized for identifying and optimizing machine learning pipelines. Afterward, he goes on to discuss regression and classification algorithms, illustrating his points with some good examples and offering step-by-step assistance throughout. In particular, I love how the results and output from TPOT are provided to the reader to know what to expect when they run the program themselves. The author also addresses parallel training, how to use TPOT in combination with Dask, and how to deploy models utilizing AWS Cloud providers, among other topics. It would be more effective, in my opinion, to provide a link to the webpage instead of describing each setting one at a time.

Overview:Machine learning can be difficult because of the many decision which need to be made about data pre-processing, algorithms to use, and parameter settings. Automated machine learning (AutoML) seeks to automate this process thus eliminating much of the guesswork from the user. This book introduces the tree-based pipeline optimization tool or TPOT as an open-source and Python based solution for AutoML. By working through this book, users will become familiar with generating machine learning models automatically using TPOT.What I like:The first chapter of the book provides a nice introduction to machine learning and sets the stage for the need for automation.Chapter 2 takes a deep dive into TPOT providing the reader with the details of the algorithms used for discovering and optimizing machine learning pipelines. I found this overview to be complete and informative. I like the use of screenshots to show the reader what they should be seeing on the screen and they install and run the software.Chapters 3 & 4 take a closer look at regression and classification problems with some good examples and step by step instructions. The format is clear and should allow the reader to follow along with their own implementation of TPOT. I particularly like how the reader is shown the results and output from TPOT so they know what to expect as they run it themselves.Chapters 5-7 cover parallelizing TPOT and using TPOT to do neural network analysis. These are both important and useful chapters with so much focus on deep learning neural networks and the use of parallel computing to speed up machine learning analyses.The final two chapters cover various aspects of deploying a TPOT model for use in practice. It covers important topics such as deploying to the cloud and graphic-user interfaces (GUI). This is a great way to end the book and the examples given are both practical and useful.What I didn’t like:While the examples are very useful for learning how to run TPOT, they do not provide as much value for using TPOT to analyze real-world big data. For example, how do you scale TPOT to data with millions of rows and columns? How do you interpret TPOT models? This might make a great volume 2 for the book.What I would like to see:It would have been nice to see a final chapter with ideas for extending or improving TPOT. Such a chapter would give computer scientists some ideas for projects advance TPOT for more complex problems. What is missing? What still needs to be done? How can the algorithms be improved?

Machine Learning Automation with TPOT – ReviewFirst, I would like to congratulate Mr. Dario Radečić, the book's author, and I am delighted to review this cutting-edge technical book as part of the Artificial Intelligence and Machine Learning series (AIML).Overview:The author kept me on the edge of my seat and reading word for word from the beginning, and the way he organized each chapter was awesome. The author went from basics to advanced level concepts with clear along with excellent examples without getting bored, and a comprehensible way of each section from Idea of Automation, Practical Classification, Regression, and Neural Networks. At one level, things went into top gear and landed on the Deployed TPOT Model in Production, which was simply outstanding. I am sure every reader will take away a substantial amount of understanding and knowledge of the Machine Learning algorithm and the power pack of the TPOT. Perfectly author mentioned that “TPOT is your data science assistant. You can use it to automate everything. boring in a data science project.”What I like:Chapter 2,3 & 4"Deep Dive into TPOT" is a particularly important chapter with respect to the title of the book. Yes! Of course, the author has covered this topic well by giving the in-detail level of installing and configuring the TPOT library for standalone Python and Anaconda. If you follow the steps carefully, you will certainly get TPOT in your environment in a very few minutes. Followed by this, the author takes us to explore various datasets and all the ML processes on a touch basis and the crystal-clear implementation of the Regression and Classification model with a reasonable volume of datasets with appropriate features and detailed step by step coding was excellent and helpful, even those who are new to ML world and knowledgeable in ML space. And TPOT came into play as the final match-winning striker and exporting the optimized pipeline new python file for further usage was impressive. Even I have tried them and enjoyed the output.Chapter 5I would say "Parallel Training with TPOT and Dask" is an extra bounce for the readers and thank you very much for bringing those topics into the book for the benefit of Data Scientists and ML Engineers. The parallelism concept in Python reminds me of the threading concept in Java programming, and the sample set of examples and code was an excellent feed for the readers. Coming into Dask library usage and its advantage over data processing with sample code is notable.Chapter 6 & 7Another milestone in this book is the Crash Course in Neural Networks, in which the author quickly covered the theory of a single neuron, the theory of a single layer, and various activation functions in a rapid way. Using neural networks to classify handwritten digits in a sample program was extraordinary, Anyone can understand the author's approach and way of implementation. And the Neural Network Classifier with TPOT as well.What I didn’t like:With respect to TPOT Model Deployment and Using the Deployed TPOT Model in Production This book's chapters appear to be power packs, but they appear rushed, and if there were more details about different model implementations, I believe something similar to the earlier deep-dive would be helpful for readers. And I believe that the author might have thought of, it would be too much for the readers, resulting in a diluting of the rhythm.What I would like to see:I was expecting the training machine learning models with TPOT and Dask libraries with a reasonable data set, which was discussed in the initial chapters, would help readers to understand how to connect the dots. And one more thing is the goal is to create a file from an automation library, It would be great if we have detailed explanation/coverage of the optimized pipeline of a new python file for the readers; if we understood this, it would be an additional value-added as a take away from this book.Overall … I can give 4.5/5 for this. And all the absolute best for the author.