Practical Machine Learning

Summary

Build useful machine learning models that deliver data-driven insights to help your company make better decisions that can improve revenue, reduce costs, create new opportunities, identify new ideas, improve the customer experience and more.

Associated project work

In this miniproject students will work with a housing data set to develop a model to predict house price based on various features about a house. They will start with a linear regression model using just one feature, then build a model with 2 features and finally use a linear regression model trained on all of the data. The miniproject requires students to be able to use scikit-learn's transformers, predictors and pipelines. To further improve the model, students will also engineer polynomial features and train a ridge model.

Students will develop a model to predict customer churn from various customer features. They will start with a logistic regression model that uses only numerical features, then they will also incorporate categorical features. Students will improve the model by using a radnom forest classifier.

Students will gain experience with implementing unsupervised learning algorithms on real-world data and drawing insightful conclusions. They will use a data set of credit card company customers and first perform principal components analysis. Then students will train a K-Means clustering model and determine the proper number of clusters according to silhouette score. Lastly they will find cluster centroids which correspond to average customers and extract characteristic customer information.

This module is currently part of our Data Science Essentials Course.

Introduction to Machine Learning

Machine learning basics Types of machine learning Training a supervised ML model Choosing a family of models Defining a cost function Finding the optimal model Conclusion

Scikit-learn API

Classes vs. objects Predictors Transformers Pipelines Conclusion

Regression

Linear regression Linear regression with a single feature Regression metrics Linear regression with multiple features Optimization Adding features Regularization Conclusion

Bias, Variance, and Overfitting

Train a decision tree model In-sample error Out-of-sample error Bias-variance tradeoff Cross-validation strategies Grid search for tuning hyperparameters Test data set: evaluating the final model Conclusion

Introduction to Classification

Customer classification Logistic regression Classification metrics: accuracy Classification metrics: precision and recall Other classification metrics Probabilistic models Multiclass classification problems Conclusion

Classification: Predicting Customer Churn

Customer data Encoding Categorical Variables One-hot encoding of all categorical columns Training a classifier Improving the model: random forest classifier Model comparison Dealing with unbalanced classes Conclusion

Dimensionality Reduction

Principal component analysis (PCA) Using real-world data PCA in practice - scaling the data Principal axes and explained variance Choosing number of components to use Interpreting the components Conclusion Appendix: PCA in detail

Clustering

K-Means clustering Metrics for clustering Choosing the number of clusters Clustering on customers data Conclusion

Scikit-learn Workflow

Remark Categorical features Combining features using ColumnTransformer Building custom scikit-learn classes Building a custom transformer using LATITUDE and LONGITUDE Using FunctionTransformer to build stateless transformers Feature unions Imputation A fancy display... Conclusion

Prerequisites

Basic Python

Basic linear algebra

Basic statistics

The Data Incubator

Practical Machine Learning

Summary

Associated project work