Kumar Alok - Practical Full Stack Machine Learning

www.bpbonline.com

FIRST EDITION 2022

Copyright BPB Publications, India

ISBN: 978-93-91030-42-1

All Rights Reserved. No part of this publication may be reproduced, distributed or transmitted in any form or by any means or stored in a database or retrieval system, without the prior written permission of the publisher with the exception to the program listings which may be entered, stored and executed in a computer system, but they can not be reproduced by the means of publication, photocopy, recording, or by any electronic and mechanical means.

LIMITS OF LIABILITY AND DISCLAIMER OF WARRANTY

The information contained in this book is true to correct and the best of authors and publishers knowledge. The author has made every effort to ensure the accuracy of these publications, but publisher cannot be held responsible for any loss or damage arising from any information in this book.

All trademarks referred to in the book are acknowledged as properties of their respective owners but BPB Publications cannot guarantee the accuracy of this information.

www.bpbonline.com

Alok is an author, speaker, open source contributor and a ML practitioner. He is currently leading the India Innovation center at Publicis Sapient to leverage emerging technologies to solve real world challenges.

He has extensive experience in leading strategic initiatives and driving cutting edge fast-paced data driven solutions ranging from products to platforms. His work has won several reputed awards. The inspiration to write the book on full-stack ML came from the observation of the struggle of scaling, productioning ML systems and teams.

Beyond work, He is passionate about democratizing knowledge. He manages multiple not-for-profit learnings and creative groups in NCR. He can be reached at linkedin (https://www.linkedin.com/in/aloksaan/) and twitter (https://twitter.com/Aloksaan).

Abhijeet Prakash has 6 years of extensive experience in Artificial Intelligence, Machine Learning and, Full Stack Development, using tools like Selenium, BS4, Google Colab, Apigee, FastAPI AWS, Google Cloud Platform with programming languages like Ruby, Node, Python, etc. Abhijeet pursued MCA from the Department of Mathematical Science and Information Technology, B.U. He has worked with various banks, NBFCs, and Fin-Tech companies. He worked as Machine Learning Engineer and currently working as a Full Stack AI Engineer. Abhijeet also wrote a book on Ethical Hacking.

I would like to thank my family for all their help, patience, and support. Without their support and assistance, I couldnt have imagined completing this book.

All the tools used in the book are coming from open source community and I would like to take this opportunity to thank the community that has helped democratize the AI knowledge so much.

Finally, I would really like to thank the team at BPB for all their help and support in my journey writing this book. It has been a pleasure to write this book and the team at BPB are certainly a big part of that. Many thanks to my development coordinator - Priyanka. Her continuous and regular follow ups helped me tremendously to keep us on track and focussed.

A successful data science project is not just about building powerful models, but the efficient execution of the entire project lifecycle. Unfortunately, data science has been made like ART and data scientist as ARTIST that uses hard to guess and unexplainable tricks.

If you find it difficult to decide on the correct initialization of hyper parameters or re-running someone else model training code, then you share the pain and frustration of data scientist community. Experience may teach you few tricks but there are limitations on how much we can remember and recall them accurately at the time of need.

Also, in the grand scheme of things, ML training is a part of a bigger data machine. This means inputs and outputs will always be reliant on other parts of the system. We would like you to pause here and think about the question How would you revert your productionized model that is failing to meet the requirements or a step before it how would you know it is failing or the accuracy has gone down the accepted limit? Again, how many such tricks across the whole pipeline would you able to learn, remember and recall?

The objective of this book is to introduce you to a collection of powerful, open

Source tools and concepts needed to build an effective data science pipeline so that you dont have to remember the tricks but only remember the right tools, which We guess and, in my experience, is much easier to do.

To ensure that you share the excitement with me consider this example.

You want to buy stocks and hence forth decided to seek advice from advisors. Being an experienced investor, you want to take the advice from others as well. Here is how it may look like. The percentage is the accuracy rate.

• Financial Advisor 75%
• Stock Market Trader 70%
• Market Research Team 75%
• Social Media Expert 60%

As clearly seen, all specialists predictions are below 75%, however if you combine all their predictions, you get a completely different picture

Accuracy Rate = 1- (25%* 30% * 25% * 40%) = 99.25%

This is the power of ensembling and in machine learning world, ensemble learning is essentially a combination of multiple machine learning techniques performed together.

Now, to experiment different ensembling techniques, you can write it from scratch, or you can use ML-Ensemble library. ML-Ensemble combines a Scikit-learn high-level API with a low-level computational graph framework to build memory efficient, maximally parallelized ensemble networks in as few lines of codes as possible. We hope this drives home the idea and purpose of this book.

Since the book is about building effective pipelines and systems, we have organized the books around common steps of data science project. The steps look like this:

Figure 0.1: CRISM DM common steps

I am sure you would have seen a picture or diagram like this before. The steps are so common that it doesnt turns any head. It seems so logical that it is hard to believe that considerable effort was spent to build this intuitive process. Interestingly, this was created before data science became the sexiest job.

CRISP DM or Cross Industry standard process for data mining is a process methodology for data mining applications.