Join our book community on Discord

https://packt.link/EarlyAccessCommunity

While ML algorithm design may not be the primary role of ML solutions architects, it is still essential for them to possess a comprehensive understanding of common real-world ML algorithms and their applications in solving business problems. This knowledge empowers ML solutions architects to identify suitable data science solutions and design the necessary technology infrastructure for deploying these algorithms effectively.By familiarizing themselves with a range of ML algorithms, ML solutions architects can grasp the strengths, limitations, and specific use cases of each algorithm. This enables them to evaluate business requirements accurately and select the most appropriate algorithmic approach to address a given problem. Whether it's classification, regression, clustering, or recommendation systems, understanding the underlying algorithms equips architects with the knowledge required to make informed decisions...

Technical requirements

You need a personal computer (Mac or Windows) to complete the hands-on exercise portion of this chapter.You need to download the dataset from https://www.kaggle.com/mathchi/churn-for-bank-customers. Additional instructions will be provided in the Hands-on exercise section.

How machines learn

In Chapter 1, "Machine Learning and Machine Learning Solutions Architecture," we discussed the self-improvement capability of ML algorithms through data processing and parameter updates, leading to the generation of models akin to compiled binaries in computer source code. But how does an algorithm actually learn? In essence, ML algorithms learn by optimizing an objective function, also known as a loss function, which involves minimizing or maximizing it. An objective function can be seen as a business metric, such as the disparity between projected and actual product sales. The aim of optimization is to reduce this disparity. To achieve this, an ML algorithm iterates and processes extensive historical sales data (training data), adjusting its internal model parameters until the gaps between projected and actual values are minimized. This process of finding the optimal model parameters is referred to as optimization, with mathematical routines specifically...

Overview of ML algorithms

The field of machine learning has seen the development of numerous algorithms, with ongoing research and innovation from both academia and industry. In this section, we will explore several well-known traditional and deep learning algorithms, examining their applications across various types of machine learning problems. Before we delve into these algorithms, it's important to discuss the factors to consider when selecting an appropriate algorithm for a given task.

Consideration for choosing ML algorithms

When choosing a machine learning algorithm, there are several key considerations to keep in mind:

• Problem type: Different algorithms are better suited for specific types of problems. For example, classification algorithms are suitable for tasks where the goal is to categorize data into distinct classes, while regression algorithms are used for predicting continuous numerical values. Understanding the problem type is crucial in selecting the most appropriate...

Hands-on exercise

In this hands-on exercise, we will build a Jupyter Notebook environment on your local machine and build and train an ML model in your local environment. The goal of the exercise is to get some familiarity with the installation process of setting up a local data science environment, and learn how to analyze the data, prepare the data, and train an ML model using one of the algorithms we covered in the preceding sections. First, let's take a look at the problem statement.

Problem statement

Before we start, let's first review the business problem that we need to solve. A retail bank has been experiencing a high customer churn rate for its retail banking business. To proactively implement preventive measures to reduce potential churn, the bank needs to know who the potential churners are, so the bank can target those customers with incentives directly to prevent them from leaving. From a business perspective, it is far more expensive to acquire a new customer...

Summary

In this chapter, we have explored various machine learning (ML) algorithms that can be applied to solve different types of ML problems. By now, you should have a good understanding of which algorithms are suitable for specific tasks. Additionally, you have set up a basic data science environment on your local machine, utilized the scikit-learn ML libraries to analyze and preprocess data, and successfully trained an ML model.In the upcoming chapter, our focus will shift to the intersection of data management and the ML life cycle. We will delve into the significance of effective data management and discuss how to build a comprehensive data management platform on AWS (Amazon Web Services) to support downstream ML tasks. This platform will provide the necessary infrastructure and tools to streamline data processing, storage, and retrieval, ultimately enhancing the overall ML workflow.