Ryan Giggs

Posted on Oct 21

Oracle Autonomous Database: Machine Learning with AutoML and Natural Language Queries with Select AI

#oraclemachinelearning #automl #selectai #nlp

Oracle Autonomous Database integrates powerful machine learning capabilities and generative AI features that democratize data science and make database interactions accessible to everyone. Oracle Machine Learning AutoML automates complex machine learning workflows, while Select AI enables natural language database queries without SQL expertise.

Oracle Machine Learning for Python (OML4Py)

OML4Py Overview

Oracle Machine Learning for Python (OML4Py) is a Python API that supports the machine learning process including data exploration and preparation, machine learning modeling, and solution deployment using your Oracle AI Database or Oracle Autonomous AI Database.

Key Capabilities:

In-Database Machine Learning: Run algorithms directly in the database
Embedded Python Execution: Execute Python functions in database-spawned engines
Automated Machine Learning (AutoML): Automated algorithm selection, feature selection, and model tuning
Scalability: Leverage database parallel processing capabilities
Data Proximity: Process data where it resides without extraction

AutoML: Automating Machine Learning Workflows

The AutoML Value Proposition

Automated Machine Learning (AutoML) enhances user productivity and machine learning results through automated algorithm and feature selection, as well as automated model tuning and selection.

AutoML automates tedious parts of data scientists' workflow and enables non-experts to produce models, democratizing machine learning across organizations.

Benefits:

Enhanced Productivity: Reduces time spent on routine ML tasks
Non-Expert Enablement: Allows business users to build models
Best Practice Automation: Incorporates data science expertise automatically
Reduced Errors: Eliminates manual mistakes in model selection
Faster Time-to-Value: Accelerates ML project delivery

AutoML Workflow Stages

The AutoML pipeline consists of six essential stages that guide data from raw input to production-ready models:

1. Dataset Preparation:

Prepare and clean your training data
Create Oracle Machine Learning DataFrame proxy objects
Separate predictors from target variables
Handle missing values and data quality issues

2. Algorithm Selection:
Identify top-k algorithms most suitable for your dataset and prediction task.

Algorithm Selection Process:

from oml import automl

# Define algorithm selection object
as_object = automl.AlgorithmSelection(
    mining_function='classification',
    score_metric='accuracy',
    parallel=2
)

# Select top algorithms
algorithm_ranking = as_object.select(X_data, y_data, k=4)

Supported Algorithms:

Support Vector Machines (SVM) - Gaussian and Linear
Neural Networks
Random Forest
Decision Trees
Logistic Regression
Naive Bayes

3. Adaptive Sampling:
Select a suitable subsample for model training to optimize performance and training time.

Sampling Benefits:

Reduces training time for large datasets
Maintains statistical significance
Oracle's advanced meta-learning intelligence suggests optimal sample sizes
Balances accuracy with computational efficiency

4. Feature Selection:
Select relevant features that contribute most significantly to prediction accuracy while reducing dimensionality.

Feature Selection Implementation:

# Define feature selection object
fs_object = automl.FeatureSelection(
    mining_function='classification',
    score_metric='accuracy',
    parallel=2
)

# Reduce features
selected_features = fs_object.reduce(
    selected_algorithm,
    X_data,
    y_data
)

Feature Selection Benefits:

Reduces model training time without compromising performance
Uses multiple feature-ranking algorithms
Oracle advanced meta-learning techniques prune search space efficiently
Handles both classification and regression problems

5. Model Tuning:
Identify optimal hyperparameters for the selected algorithm to maximize model performance.

Model Tuning Process:

# Define model tuning object
mt_object = automl.ModelTuning(
    mining_function='classification',
    score_metric='accuracy',
    parallel=2
)

# Tune model
tuned_model = mt_object.tune(
    selected_algorithm,
    X_reduced,
    y_data
)

Hyperparameter Optimization:

Automated parameter search across algorithm-specific hyperparameter space
Cross-validation for robust performance estimation
Grid search and random search strategies
Oracle meta-learning for intelligent parameter recommendations

6. Feature Prediction Impact:
Compute the impact of each input column on the predictions of the final tuned model.

Impact Analysis Benefits:

Provides insights into tuned AutoML model behavior
Helps business users understand model decisions
Identifies most influential features
Supports model explainability and transparency

AutoML UI: Visual Model Development

No-Code ML Development:
To use the Oracle Machine Learning AutoML UI, you start by creating an experiment. An experiment is a unit of work that minimally specifies the data source, prediction target, and prediction type.

AutoML UI Workflow:

Create Experiment: Define data source, target, and prediction type
Configure Settings: Set maximum models, run duration, metrics
Start Experiment: Choose between "Faster Results" or "Better Accuracy"
Review Results: Examine ranked models by quality metric
Deploy or Generate Notebook: Deploy models or generate Python code

Experiment Configuration:

Maximum Top Models: Select number of top models to create (default 5)
Maximum Run Duration: Set time limit (default 8 hours)
Database Service Level: Configure query parallelism
Model Metric: Choose evaluation metric (accuracy, F1, precision, recall)
Algorithm Selection: Select specific algorithms or use all candidates

After an experiment runs successfully, it presents you a list of machine learning models in order of model quality according to the metric selected.

Deployment Options:

Direct Deployment: Deploy selected models for scoring
Notebook Generation: Generate Python code using OML4Py and specific settings AutoML used to produce the model
Model Comparison: Compare models across multiple metrics
Model Explanation: Understand feature importance and prediction drivers

AutoML API: Programmatic Model Development

OML4Py AutoML Classes:

automl.AlgorithmSelection: Rank algorithms by predicted score
automl.FeatureSelection: Identify most relevant feature subsets
automl.ModelTuning: Tune hyperparameters for specific algorithms
automl.ModelSelection: Select top algorithm and tune automatically

Model Selection Example:

from oml import automl

# Automatic model selection and tuning
ms_object = automl.ModelSelection(
    mining_function='classification',
    score_metric='f1_weighted',
    parallel=4
)

# Select and tune best model
best_model = ms_object.select(X_train, y_train)

# Score on test data
predictions = best_model.predict(X_test)

Select AI: Natural Language Database Queries

Select AI Overview

Select AI makes it easy to get answers about your business using natural language, transforming how users interact with databases by eliminating the need for SQL expertise.

How Select AI Works:
Select AI processes questions using an AI Large Language Model (LLM), translating natural language prompts into SQL queries that the database executes automatically.

The DBMS_CLOUD_AI package enables integration with a user-specified LLM for generating SQL code using natural language prompts.

Select AI Process:

User Provides Natural Language Prompt: "What were our top 5 products by revenue last quarter?"
Database Augments Prompt: Adds schema metadata to provide context
LLM Generates SQL: Translates augmented prompt to SQL query
Database Executes Query: Runs generated SQL and returns results
Natural Language Response: Optionally narrates results in natural language

Key Select AI Features

Simple and Intuitive:

Use plain language instead of complex SQL syntax
No need to understand database schema or table relationships
Multilingual support for global businesses
Voice command support for hands-free queries

Future-Enabled:

Integration with cutting-edge LLM technologies
Continuous improvement as LLMs evolve
Support for emerging AI capabilities
Regular feature enhancements and updates

Secure:

Respects existing database security policies
Access control based on user permissions
Data never leaves your organization
Compliant with data governance requirements

Easy to Extend:

Build new natural language processing applications
Integrate with existing tools and workflows
Develop custom AI-driven database interactions
API support for programmatic access

Supported LLM Providers

Select AI supports multiple Large Language Model providers, giving organizations flexibility to choose their preferred AI platform:

OCI Generative AI:

Oracle's native generative AI service
Models from Meta (Llama) and Cohere hosted on OCI
Strong data governance and security
Custom models trained on customer data
Region-specific deployment options

OpenAI:

GPT-3.5 and GPT-4 models
Advanced natural language understanding
Broad language support
Regular model updates and improvements

Cohere:

Enterprise-focused LLMs
Specialized text generation and understanding
Multilingual capabilities
Privacy-preserving features

Azure OpenAI:

Microsoft's enterprise OpenAI service
GPT models with Azure integration
Enterprise security and compliance
Hybrid cloud deployment options

Select AI Actions

Natural Language to SQL:
Generate and execute SQL queries from natural language prompts.

SELECT AI what were the top 5 selling products last month;

SHOWSQL - View Generated Query:
Display the SQL query generated from natural language without executing it.

SELECT AI SHOWSQL what are the top 5 products by revenue;

NARRATE - Natural Language Results:
Execute query and return results as natural language description instead of tabular data.

SELECT AI NARRATE what were our total sales by region last quarter;

CHAT - General AI Conversation:
Submit general requests and questions to the LLM, with optional context from vector stores.

SELECT AI CHAT explain the concept of customer lifetime value;

Retrieval Augmented Generation (RAG)

Select AI with RAG:
Select AI with RAG augments your natural language prompt by retrieving content from your specified vector store using semantic similarity search. This reduces hallucinations by using your specific and up-to-date content and provides more relevant natural language responses.

RAG Benefits:

Reduced Hallucinations: LLM uses your actual data and documents
Current Information: Access to latest business content
Improved Accuracy: Context-specific responses
Domain Expertise: Incorporate organizational knowledge

AI Vector Search:
AI Vector Search enables similarity queries on text, documents, images, and other unstructured data represented as vectors, powering RAG implementations.

Select AI Profile Configuration

Creating an AI Profile:

BEGIN
    DBMS_CLOUD_AI.CREATE_PROFILE(
        profile_name => 'my_ai_profile',
        attributes => JSON_OBJECT(
            'provider' VALUE 'oci',
            'credential_name' VALUE 'OCI_CRED',
            'object_list' VALUE JSON_ARRAY(
                JSON_OBJECT(
                    'owner' VALUE 'ADMIN',
                    'name' VALUE 'CUSTOMERS'
                ),
                JSON_OBJECT(
                    'owner' VALUE 'ADMIN',
                    'name' VALUE 'ORDERS'
                )
            )
        )
    );
END;
/

Profile Components:

Provider: LLM provider (OCI, OpenAI, Cohere, Azure)
Credentials: Authentication for LLM access
Object List: Schemas, tables, and views available for queries
Settings: Model-specific parameters and configurations

Select AI Use Cases

Business Intelligence:

Sales performance analysis
Customer behavior insights
Market trend identification
Financial reporting and analysis

Data Exploration:

Ad-hoc data analysis by business users
Quick insights without waiting for IT
Exploratory data analysis for new datasets
Validation of business hypotheses

Application Development:

Build conversational AI applications
Integrate natural language interfaces into existing apps
Develop chatbots with database access
Enable voice-activated database queries

Citizen Data Science:

Empower non-technical users to analyze data
Self-service analytics for departments
Reduce IT bottlenecks for simple queries
Democratize data access across organization

Integration: AutoML and Select AI

Complementary Capabilities

AutoML for Model Building:
Use AutoML to build predictive models from your data, automating the machine learning workflow.

Select AI for Model Insights:
Use Select AI to query model predictions and understand model behavior through natural language.

Combined Workflow:

Use AutoML to build customer churn prediction model
Deploy model in database
Use Select AI to query: "Which customers are most likely to churn next month?"
Get natural language explanation of predictions and contributing factors

End-to-End AI/ML Platform

Oracle Autonomous Database provides:

Integrated machine learning platform (OML4Py, OML4R, OML4SQL)
Automated model development (AutoML UI and API)
Natural language data access (Select AI)
Vector database capabilities (AI Vector Search)
Generative AI integration (LLM providers)
Scalable in-database processing

Best Practices

AutoML Best Practices

Data Preparation:

Clean and validate data before AutoML
Handle missing values appropriately
Ensure sufficient training data
Balance classes for classification problems

Experiment Configuration:

Start with "Faster Results" for quick validation
Use "Better Accuracy" for production models
Set reasonable time limits based on dataset size
Monitor resource consumption during training

Model Evaluation:

Evaluate multiple top models, not just the best
Test models on holdout data
Consider business metrics alongside statistical metrics
Validate model explainability and fairness

Select AI Best Practices

Profile Configuration:

Include only relevant schemas and tables in profiles
Use descriptive table and column comments
Maintain accurate database metadata
Create separate profiles for different use cases

Prompt Engineering:

Be specific and clear in natural language prompts
Provide context when needed
Start simple and refine based on results
Use SHOWSQL to understand generated queries

Security and Governance:

Leverage existing database security policies
Monitor LLM usage and costs
Review generated SQL for sensitive operations
Implement appropriate access controls

Quality Assurance:

Validate generated SQL queries
Review LLM responses for accuracy
Test with various prompt formulations
Implement guardrails for production use

Conclusion

Oracle Autonomous Database's integration of Oracle Machine Learning AutoML and Select AI represents a significant advancement in making advanced analytics accessible to all users. AutoML automates tedious parts of data scientists' workflow and enables non-experts to produce models, while Select AI makes it easy to get answers about your business using natural language.