6. Building Your Own Program

Overview

In this chapter, we will present all the steps required to solve a problem using machine learning. We will take a look at the key stages involved in building a comprehensive program. We will save a model in order to get the same results every time it is run and call a saved model to use it for predictions on unseen data. By the end of this chapter, you will be able to create an interactive version of your program so that anyone can use it effectively.

Introduction

In the previous chapters, we covered the main concepts of machine learning, beginning with the distinction between the two main learning approaches (supervised and unsupervised learning), and then moved on to the specifics of some of the most popular algorithms in the data science community.

This chapter will talk about the importance of building complete machine learning programs, rather than just training models. This will involve taking the models to the next level, where they can be accessed and used easily.

We will do this by learning how to save a trained model. This will allow the best performing model to be loaded in order to make predictions over unseen data. We will also learn the importance of making a saved model available through platforms where users can easily interact with it.

This is especially important when working in a team, either for a company or for research purposes, as it allows all members of the team to use the model without needing...

Program Definition

The following section will cover the key stages required to construct a comprehensive machine learning program that allows easy access to the trained model so that we can perform predictions for all future data. These stages will be applied to the construction of a program that allows a bank to determine the promotional strategy for a financial product in its marketing campaign.

Building a Program – Key Stages

At this point, you should be able to pre-process a dataset, build different models using training data, and compare those models in order to choose the one that best fits the data at hand. These are some of the processes that are handled during the first two stages of building a program, which ultimately allows the creation of the model. Nonetheless, a program should also consider the process of saving the final model, as well as the ability to perform quick predictions without the need for coding.

The processes that we just discussed are...

Saving and Loading a Trained Model

Although the process of manipulating a dataset and training the right model is crucial for developing a machine learning project, the work does not end there. Knowing how to save a trained model is key as this will allow you to save the hyperparameters, as well as the values for the weights and biases of your final model, so that it remains unchanged when it is run again.

Moreover, after the model has been saved to a file, it is also important to know how to load the saved model in order to use it to make predictions on new data. By saving and loading a model, we allow for the model to be reused at any moment and through many different means.

Saving a Model

The process of saving a model is also called serialization, and it has become increasingly important due to the popularity of neural networks that use many parameters (weights and biases) that are randomly initialized every time the model is trained, as well as due to the introduction...

Interacting with a Trained Model

Once the model has been created and saved, it is time for the last step of building a comprehensive machine learning program: allowing easy interaction with the model. This step not only allows the model to be reused, but also introduces efficiency to the implementation of machine learning solutions by allowing you to perform classifications using just input data.

There are several ways to interact with a model, and the decision that's made between choosing one or the other depends on the nature of the user (the individuals that will be making use of the model on a regular basis). Machine learning projects can be accessed in different ways, some of which require the use of an API, an online or offline program (application), or a website.

Moreover, once the channel is defined based on the preference or expertise of the users, it is important to code the connection between the final user and the model, which could be either a function or a...

Summary

This chapter wraps up all of the concepts and techniques that are required to successfully train a machine learning model based on training data. In this chapter, we introduced the idea of building a comprehensive machine learning program that not only accounts for the stages involved in the preparation of the dataset and creation of the ideal model, but also the stage related to making the model accessible for future use, which is accomplished by carrying out three main processes: saving the model, loading the model, and creating a channel that allows users to easily interact with the model and obtain an outcome.

For saving and loading a model, the pickle module was introduced. This module is capable of serializing the model to save it in a file, while also being capable of deserializing it to make use of the model in the future.

Furthermore, to make the model accessible to users, the ideal channel (for example, an API, an application, a website, or a form) needs to...