You're reading from Practical Data Science with Python Learn tools and techniques from hands-on examples to extract insights from data

Product type Paperback

Published in Sep 2021

Publisher Packt

ISBN-13 9781801071970

Length 620 pages

Edition 1st Edition

Languages

Python

Tools

Excel

Concepts

Data Science

Author (1):

Nathan George

View More author details

Table of Contents (30) Chapters

Preface

1. Part I - An Introduction and the Basics

2. Introduction to Data Science FREE CHAPTER

3. Getting Started with Python

4. Part II - Dealing with Data

5. SQL and Built-in File Handling Modules in Python

6. Loading and Wrangling Data with Pandas and NumPy

7. Exploratory Data Analysis and Visualization

8. Data Wrangling Documents and Spreadsheets

9. Web Scraping

10. Part III - Statistics for Data Science

11. Probability, Distributions, and Sampling

12. Statistical Testing for Data Science

13. Part IV - Machine Learning

14. Preparing Data for Machine Learning: Feature Selection, Feature Engineering, and Dimensionality Reduction

15. Machine Learning for Classification

16. Evaluating Machine Learning Classification Models and Sampling for Classification

17. Machine Learning with Regression

18. Optimizing Models and Using AutoML

19. Tree-Based Machine Learning Models

20. Support Vector Machine (SVM) Machine Learning Models

21. Part V - Text Analysis and Reporting

22. Clustering with Machine Learning

23. Working with Text

24. Part VI - Wrapping Up

25. Data Storytelling and Automated Reporting/Dashboarding

26. Ethics and Privacy

27. Staying Up to Date and the Future of Data Science

28. Other Books You May Enjoy

29. Index

Evaluating classification algorithm performance with metrics

There are several metrics available for comparing classification ML algorithms, each with their strengths and weaknesses. We'll look at some of the more common metrics for classification here.

Train-validation-test splits

When evaluating the performance of an algorithm, it's important to look at performance on data that was not used for training. The model has already learned all the ins and outs of the training data, and may have even overfit to the training data, learning patterns related to noise in the data. Instead, we want to evaluate performance on a hold-out set, which we may call a validation or test set. For some algorithms, such as neural networks, we train the model on a training set, monitor performance during training on a validation set, and then evaluate final performance on a test set. For most other ML algorithms, we simply use a training and test set If our test or validation sets contain...