Simplifying AI/ML Embedding with Oracle Machine Learning

Table of Contents

1. Introduction
2. Why Use Machine Learning Models
3. Where Do Models Come From
4. The Machine Learning Modeling Process
5. Embedding AI and ML in Your Applications
6. Deploying In-Database Models
7. Deploying Models to OML Services
8. Deploying R and Python-Based Solutions
9. Deploying Third-Party Models
10. Getting Started with OML on Autonomous Database

Introduction

The integration of Artificial Intelligence (AI) and Machine Learning (ML) technologies in applications has become increasingly popular among enterprises looking to enhance business operations and decision-making processes. However, deploying and scaling ML models in production applications can be a challenging task. In this article, we will explore how Oracle Machine Learning (OML) simplifies the embedding of AI and ML in applications.

Why Use Machine Learning Models

Machine learning models offer a wide range of possibilities for solving business problems by analyzing large volumes of data. Some common use cases include:

• Classification: Identifying customers with high lifetime value, predicting customer churn, and addressing HR functions like hiring and employee retention.
• Regression: Forecasting product demand and sales or revenue, predicting house prices, and identifying fraudulent claims.
• Time Series Analysis: Analyzing data over time to make predictions and identify Patterns.
• Clustering: Segmenting customers, document classification, and exploratory data analysis.
• Association Rules: Identifying cross-sell and upsell opportunities.
• Anomaly Detection: Detecting transaction or claims fraud, and identifying unusual cases.

Each use case can be addressed using multiple ML techniques, and Oracle Machine Learning supports a range of algorithms for each technique.

Pros:

• ML models provide valuable insights and recommendations based on data analysis.
• They can be used to generate predictions and make informed decisions in various applications.
• Models can be customized based on specific business needs and requirements.

Cons:

• Building and deploying ML models can be challenging, especially for non-experts.
• The selection of appropriate algorithms and hyperparameters can be time-consuming and complex.

Where Do Models Come From

ML models are created using algorithms that operate on data to generate insights, recommendations, and predictions. There are multiple tools and platforms available for building ML models:

1. Oracle Machine Learning: OML provides powerful in-database algorithms and tools directly within Oracle Database and Autonomous Database instances. These algorithms can be accessed using SQL, R, Python, and REST APIs.

2. Open Source Tools: Popular open-source tools like Python and R have rich ecosystems of third-party packages that expand their functionality across multiple domains. Oracle Database and Autonomous Database offer advantages over other tools by eliminating access latency, improving data security and governance, and providing scalability and performance optimizations.

3. Automated Machine Learning (AutoML): OML includes a no-code user interface for automated modeling. Users can specify their data and target variables, and AutoML handles the selection of algorithms, hyperparameter tuning, and model evaluation to simplify the modeling process. AutoML generates editable starter notebooks that can be customized and augmented based on domain expertise.

Pros:

• Oracle Machine Learning provides powerful in-database algorithms and tools for building ML models.
• Open-source tools like Python and R offer a wide range of packages and libraries for ML.
• Automated Machine Learning simplifies the modeling process for non-experts.

Cons:

• Building ML models requires expertise and knowledge of algorithms and hyperparameters.
• Accessing and managing data for modeling can be complex and time-consuming.

The Machine Learning Modeling Process

The machine learning modeling process involves several steps, including:

1. Data Preparation: Assembling, integrating, and transforming data to meet the requirements of the ML project.
2. Algorithm Selection: Choosing the most suitable algorithm(s) for a given use case.
3. Model Building: Building and evaluating ML models using selected algorithms and hyperparameters.
4. Model Deployment: Making the ML model available for use in applications and dashboards.
5. Model Monitoring: Periodically evaluating the performance of deployed models and updating them if necessary.

Automated Machine Learning (AutoML) simplifies and automates many of these steps, eliminating repetitive and time-consuming tasks.

Pros:

• The modeling process allows for customization and optimization of ML models.
• Automated Machine Learning reduces the time and effort required for model building.
• Regular model monitoring ensures that deployed models remain accurate and effective.

Cons:

• The modeling process can be complex and time-consuming, especially for large datasets.
• Manual tuning of hyperparameters may be required for optimal model performance.

Embedding AI and ML in Your Applications

Embedding AI and ML in applications involves the integration of ML models and insights generated from data analysis. There are multiple ways to achieve this:

1. Deploying In-Database Models: In-database ML models can be used directly in SQL queries to make predictions and generate insights. This approach eliminates the need to extract data from the database and improves data security and governance.

2. Deploying Models to OML Services: In-database models can be deployed to Oracle Machine Learning (OML) Services, which provide a REST API for flexible application integration. Scoring with OML Services is optimized for real-time and streaming applications, with fast response times.

3. Deploying R and Python-Based Solutions: R and Python solutions can be deployed within the database environment, using native interfaces to access in-database algorithms. Embedded execution allows for scalable and efficient processing of large volumes of data.

4. Deploying Third-Party Models: Models produced using third-party tools, like TensorFlow or PyTorch, can be exported in ONNX format and imported to OML Services. This approach allows for the integration of models from various platforms and frameworks.

Pros:

• Embedding AI and ML in applications enhances decision-making and provides insights for improved business operations.
• Different deployment options cater to a variety of application development preferences and requirements.
• In-database models and OML Services offer optimized performance and scalability.

Cons:

• Embedding AI and ML requires expertise in application development and integration.
• Continuous monitoring and maintenance of deployed models may be necessary to ensure accuracy and effectiveness.

Deploying In-Database Models

In-database models are ML models that are built and stored directly in the database. These models can be used within SQL queries to make predictions and generate insights. Deploying in-database models offers several advantages:

1. Reduced Access Latency: In-database models eliminate the need to extract data from the database, reducing access latency and improving overall performance.

2. Improved Data Security and Governance: Since the models are stored within the database, they inherit the security and governance features of the database, ensuring data privacy and compliance.

3. Scalability and Performance: In-database algorithms are designed to take AdVantage of memory optimizations and distributed processing capabilities, allowing for scalable and efficient model scoring.

4. Simplified Management: Storing models in the database eliminates the need to manage additional tools separately and reduces the complexity of the overall ML workflow.

To deploy in-database models, users can utilize Oracle Machine Learning (OML) functions and interfaces, such as OML for SQL, OML for R, and OML for Python. These interfaces provide access to a wide range of algorithms and tools directly within the database environment.

Pros:

• Reduced access latency improves overall performance.
• Data security and governance are enhanced within the database environment.
• Scalable and efficient model scoring is supported.

Cons:

• Deploying in-database models may require expertise in SQL and database management.
• Limited flexibility compared to models deployed using other methods.

Deploying Models to OML Services

OML Services provide a REST API for deploying and managing machine learning models. This approach enables flexible application integration and optimized model scoring for real-time and streaming applications. Deploying models to OML Services offers several benefits:

1. Flexible Application Integration: Models deployed to OML Services can be accessed and scored using REST endpoints, making it easy to integrate ML capabilities into various applications and frameworks.

2. Scalability and Performance: OML Services are provisioned and maintained as part of Oracle Autonomous Database, ensuring scalability and optimal performance for model scoring.

3. Reduced Cost: Users only pay for the compute resources used when producing predictions with OML Services, leading to cost savings compared to provisioning separate VMs for 24/7 availability.

4. Model Management and Monitoring: OML Services enable the deployment of models from both Oracle Database and Autonomous Database, as well as third-party models exported in ONNX format. Models can be monitored for concept drift and changes in quality metrics.

To deploy models to OML Services, users can utilize the OML Models interface in Oracle Machine Learning. This interface provides tools for managing and deploying models, as well as monitoring model performance and data quality.

Pros:

• Flexibility in application integration through REST API endpoints.
• Scalability and optimal performance for model scoring.
• Cost-effective deployment and management of ML models.

Cons:

• Deployment to OML Services may require additional configuration and management.
• Integration with existing applications may require development effort.

Deploying R and Python-Based Solutions

Deploying R and Python-based solutions involves integrating custom scripts and functions written in R and Python into the database environment. This approach allows for the utilization of third-party packages and libraries and enables scalable and efficient processing of data. Deploying R and Python-based solutions offers several advantages:

1. Access to Rich Ecosystems: R and Python have vast ecosystems of packages and libraries that expand the functionality of the language across multiple domains. This allows for the utilization of specialized algorithms and techniques in ML modeling.

2. Scalable and Efficient Processing: R and Python solutions can be executed within the database environment, leveraging the scalability and performance optimizations of the database. This enables the processing of large volumes of data in a distributed and Parallel manner.

3. Seamless Integration: Custom R and Python scripts and functions can be seamlessly integrated into the database environment, making them easily accessible for application development using SQL, REST endpoints, or native interfaces.

To deploy R and Python-based solutions, users can utilize Oracle Machine Learning (OML) interfaces, such as OML for R and OML for Python. These interfaces provide native R and Python execution environments within the database environment, enabling the execution of custom scripts and functions.

Pros:

• Access to a rich ecosystem of packages and libraries in R and Python.
• Scalable and efficient processing of large volumes of data.
• Seamless integration with the database environment.

Cons:

• Development and integration of custom R and Python scripts may require expertise in R, Python, and database management.
• Managing dependencies and version control can be challenging.

Deploying Third-Party Models

OML Services support the deployment of third-party ML models in ONNX format. This allows users to leverage models built using popular frameworks like TensorFlow and PyTorch. Deploying third-party models offers several benefits:

1. Integration of Various Models: Users can export ML models built in popular frameworks as ONNX files and import them into OML Services. This allows for the integration of models across different platforms and frameworks.

2. Seamless Application Integration: Models deployed in OML Services can be accessed and scored using standard REST API endpoints. This makes it easy to integrate third-party models into various applications and frameworks.

3. Platform Flexibility: Users can choose the most suitable ML framework for their tasks and workflows. OML Services provide a unified platform for deploying and managing models, regardless of the framework used.

To deploy third-party models to OML Services, users need to export the models in ONNX format and import them into the services using the REST API. Once deployed, the models can be accessed and scored using REST endpoints.

Pros:

• Integration of models built using popular ML frameworks.
• Seamless application integration through REST API endpoints.
• Flexibility in choosing ML frameworks.

Cons:

• Exporting models in ONNX format may require additional steps and configuration.
• Integration with existing applications may require development effort.

Getting Started with OML on Autonomous Database

To get started with Oracle Machine Learning (OML) on Autonomous Database, follow these steps:

1. Sign into your Oracle Autonomous Database instance as an admin user.
2. In the database actions interface, scroll down to Administration and click on "Database Users" to Create a new OML user.
3. Specify the user's name, password, and enable OML and web access for the user.
4. Sign in to the OML user interface using the newly created user account.
5. Explore the OML tools and features, such as the automl UI, notebooks, jobs, and examples.

By following these steps, users can access and utilize the various components and functionalities of OML on Autonomous Database.

Pros:

• Easy setup and access to OML tools and features.
• Integration with Oracle Autonomous Database for improved performance and scalability.
• Access to a wide range of tools and algorithms for ML modeling.

Cons:

• Requires administrative access to Oracle Autonomous Database.
• Familiarity with SQL, R, and Python is necessary for optimal use of OML.

Conclusion

Oracle Machine Learning (OML) simplifies the embedding of AI and ML in applications by providing powerful algorithms, tools, and services within Oracle Database and Autonomous Database instances. With features like in-database models, OML Services, and integration with R and Python-based solutions, developers can easily build and deploy ML models to enhance their applications' capabilities. By following the steps outlined in this article, users can explore and leverage OML to harness the power of AI and ML in their own applications.

Highlights

• Oracle Machine Learning (OML) simplifies the embedding of AI and ML in applications.
• ML models can be built using in-database algorithms, R, Python, and automated modeling.
• Deploying in-database models and using OML Services provide optimized performance and scalability.
• R and Python-based solutions can be seamlessly integrated into the database environment.
• OML supports the deployment of third-party models built with popular frameworks.
• Getting started with OML on Autonomous Database is easy and straightforward.

FAQ

Q: What are the advantages of deploying in-database models?

A: Deploying in-database models eliminates access latency, improves data security and governance, ensures scalability and performance optimization, and simplifies model management within the database environment.

Q: Can I use Python and R packages in my ML models deployed to Oracle Database?

A: Yes, Oracle Machine Learning (OML) supports the embedding of Python and R packages, allowing users to leverage the rich ecosystems of these languages in their ML models.

Q: Are there any additional costs associated with deploying models to OML Services?

A: No, deploying models to OML Services is included in the database license or subscription cost. Users only pay for the additional compute resources used when producing predictions with OML Services.

Q: Can I integrate models built with other ML frameworks, such as TensorFlow or PyTorch, into OML Services?

A: Yes, OML Services support the deployment of ML models in ONNX format, allowing users to leverage models built with popular frameworks like TensorFlow and PyTorch.

Q: Is it possible to access and score in-database models using REST endpoints?

A: In-database models can be accessed and scored using SQL queries. However, if You want to access them through REST endpoints, you can deploy the models to OML Services and utilize the REST API provided.

Q: How can I get started with Oracle Machine Learning on Autonomous Database?

A: To get started with Oracle Machine Learning on Autonomous Database, follow the steps outlined in the "Getting Started with OML on Autonomous Database" section of this article.