book

Low-Code AI

by Gwendolyn Stripling, Michael Abel

September 2023

Intermediate to advanced

328 pages

8h 47m

English

O'Reilly Media, Inc.

Book available

Read now

Unlock full access

Includes

Includes Quizzes

Who Should Read This Book?What Is and Isn’t in This BookConventions Used in This BookUsing Code ExamplesO’Reilly Online LearningHow to Contact UsAcknowledgments
What Is the Goal or Use Case?An Enterprise ML WorkflowDefining the Business Objective or Problem StatementData CollectionData PreprocessingData AnalysisData Transformation and Feature SelectionResearching the Model Selection or Using AutoML (a No-Code Solution)Model Training, Evaluation, and TuningModel TestingModel Deployment (Serving)Maintaining ModelsSummary
Overview of Use Cases and Datasets Used in the Book1. Retail: Product Pricing2. Healthcare: Heart Disease Campaign3. Energy: Utility Campaign4. Insurance: Advertising Media Channel Sales Prediction5. Financial: Fraud Detection6. Energy: Power Production Prediction7. Telecommunications: Customer Churn Prediction8. Automotive: Improve Custom Model PerformanceData and File TypesQuantitative and Qualitative DataStructured, Unstructured, and Semistructured DataData File TypesHow Data Is ProcessedAn Overview of GitHub and Google’s ColabUse GitHub to Create a Data Repository for Your ProjectsUsing Google’s Colaboratory for Low-Code AI ProjectsSummary
No-Code AutoMLHow AutoML WorksMachine Learning as a ServiceLow-Code ML FrameworksSQL ML FrameworksOpen Source ML LibrariesSummary
The Business Use Case: Media Channel Sales PredictionProject WorkflowProject DatasetExploring the Dataset Using Pandas, Matplotlib, and SeabornLoad Data into a Pandas DataFrame in a Google Colab NotebookExplore the Advertising DatasetUse AutoML to Train a Linear Regression ModelNo-Code Using Vertex AICreate a Managed Dataset in Vertex AISelect the Model ObjectiveBuild the Training ModelEvaluate Model PerformanceModel Feature Importance (Attribution)Get Predictions from Your ModelSummary
The Business Use Case: Fraud Detection for Financial TransactionsProject WorkflowProject DatasetExploring the Dataset Using Pandas, Matplotlib, and SeabornLoading Data into a Pandas DataFrame in a Google Colab NotebookExploring the DatasetExporting the DatasetClassification Models and MetricsUsing AutoML to Train a Classification ModelCreating a Managed Dataset and Selecting the Model ObjectiveExploring Dataset StatisticsTraining the ModelEvaluating Model PerformanceModel Feature ImportancesGetting Predictions from Your ModelSummary
The Business Use Case: Power Plant ProductionCleaning the Dataset Using SQL in BigQueryLoading a Dataset into BigQueryExploring Data in BigQuery Using SQLLinear Regression ModelsFeature Selection and CorrelationGoogle ColaboratoryPlotting Feature Relationships to the LabelThe CREATE MODEL Statement in BigQuery MLIntroducing Explainable AIExplainable AI in BigQuery MLExercisesNeural Networks in BigQuery MLBrief Overview of Neural NetworksActivation Functions and NonlinearityTraining a Deep Neural Network in BigQuery MLExercisesDeep Dive: Using Cloud Shell to View Your Cloud Storage FileSummary
The Business Use Case: Customer Churn PredictionChoosing Among No-Code, Low-Code, or Custom Code ML SolutionsExploring the Dataset Using Pandas, Matplotlib, and SeabornLoading Data into a Pandas DataFrame in a Google Colab NotebookUnderstanding and Cleaning the Customer Churn DatasetTransforming Features Using Pandas and Scikit-LearnBuilding a Logistic Regression Model Using Scikit-LearnLogistic RegressionTraining and Evaluating a Model in Scikit-LearnClassification Evaluation MetricsServing Predictions with a Trained Model in Scikit-LearnPipelines in Scikit-Learn: An IntroductionBuilding a Neural Network Using KerasIntroduction to KerasTraining a Neural Network Classifier Using KerasBuilding Custom ML Models on Vertex AISummary
The Business Use Case: Used Car Auction PricesModel Improvement in Scikit-LearnLoading the Notebook with the Preexisting ModelLoading the Datasets and the Training-Validation-Test Data SplitExploring the Scikit-Learn Linear Regression ModelFeature Engineering and Improving the Preprocessing PipelineHyperparameter TuningModel Improvement in KerasIntroduction to Preprocessing Layers in KerasCreating the Dataset and Preprocessing Layers for Your ModelBuilding a Neural Network ModelHyperparameter Tuning in KerasHyperparameter Tuning in BigQuery MLLoading and Transforming Car Auction DataTraining a Linear Regression Model and Using the TRANSFORM ClauseConfigure a Hyperparameter Tuning Job in BigQuery MLOptions for Hyperparameter Tuning Large ModelsVertex AI Training and TuningAutomatic Model Tuning with Amazon SageMakerAzure Machine LearningSummary
Going Deeper into Data ScienceWorking with Unstructured DataGenerative AIExplainable AIML OperationsContinuous Training and EvaluationSummary

Content preview from Low-Code AI

Chapter 5. Using AutoML to Detect Fraudulent Transactions

In this chapter, you build a Vertex AI AutoML model to predict whether a financial transaction is fraudulent or not. You will clean and explore the dataset in a Google Colab notebook environment before creating a managed dataset on Vertex AI as you did in Chapter 3. Once you have created a managed dataset, you will use AutoML to create a classification model to predict if a transaction is fraudulent or not. Along the way, the chapter discusses classification models in general and the corresponding metrics that are commonly used to evaluate them.

The overall workflow of this chapter is very similar to what you worked through in Chapter 4 for the problem of predicting advertising media channel sales. For this reason, in many places in this chapter you will see more concise details where the conversations would be very similar. If you get stuck in these sections, please refer back to Chapter 4 for more details.

The Business Use Case: Fraud Detection for Financial Transactions

Your task in this chapter, as mentioned, is to build a model to predict whether a financial transaction is fraudulent or legitimate. Your new company is a mobile payment service that serves hundreds of thousands of users. Fraudulent transactions are fairly rare and are usually caught by other protections. However, the unfortunate truth is that some of these are slipping through the cracks and negatively impacting your users. Your company can rectify ...