td-change-point

name	td-change-point
description	Change point detection in time series for structural breaks

Teradata Change Point Detection

Skill Name	Teradata Change Point Detection
Description	Change point detection in time series for structural breaks
Category	Time Series Analysis
Function	ChangePointDetection
Framework	Teradata Machine Learning Engine (MLE)

Core Capabilities

• MLE table-operator implementation with PARTITION BY and ORDER BY support
• Scalable time series analysis for millions of products or billions of IoT sensors
• Multiple segmentation methods including normal distribution modeling
• Production-ready SQL generation with proper MLE ON...USING syntax
• Comprehensive error handling and data validation
• Business-focused interpretation of analytical results
• Flexible search methods (binary, linear) for change point detection

Machine Learning Engine (MLE) Overview

The Machine Learning Engine is Teradata's framework for advanced analytics using table operators:

• Table-operator pattern: SELECT * FROM Function(ON table PARTITION BY ... ORDER BY ... USING ...)
• Parallel execution across AMPs via PARTITION BY
• Built-in functions for classification, regression, time series, and more
• SQL-native integration with standard Teradata workflows

Table Analysis Workflow

This skill automatically analyzes your time series data to generate optimized MLE workflows:

1. Time Series Structure Analysis

• Temporal Column Detection: Identifies time/date columns for ordering
• Value Column Classification: Distinguishes between numeric time series values
• Series ID Detection: Identifies grouping columns for PARTITION BY
• Frequency Analysis: Determines sampling frequency and intervals

2. MLE-Specific Recommendations

• Partition Strategy: Configures PARTITION BY for parallel processing
• Parameter Optimization: Suggests optimal parameters for ChangePointDetection
• Search Method Selection: Recommends binary vs linear search
• Segmentation Method: Suggests appropriate statistical model

3. SQL Generation Process

• MLE Syntax Generation: Creates proper ON...PARTITION BY...ORDER BY...USING SQL
• Parameter Configuration: Sets function-specific parameters
• Result Storage: Generates CREATE TABLE AS patterns for persisting results

How to Use This Skill

1. Provide Your Time Series Data:

   "Analyze time series table: database.sensor_data with timestamp column and value columns"
   

2. The Skill Will:

   • Analyze temporal structure and sampling frequency
   • Identify optimal function parameters
   • Generate complete ChangePointDetection workflow
   • Provide performance optimization recommendations

Input Requirements

Data Requirements

• Time series table: Teradata table with temporal data
• Time column: Time/date column for ORDER BY
• Value column: Numeric column to analyze for change points
• Series ID column: Column to PARTITION BY (grouping key for parallel processing)

Technical Requirements

• Teradata Vantage with Machine Learning Engine (MLE) enabled
• MLE License: Access to analytic table operators
• Database permissions: CREATE, DROP, SELECT on working database
• Function access: ChangePointDetection

Output Formats

Generated Results

• Change point locations with timestamps
• Segment boundaries identifying structural breaks
• Statistical metrics for each detected change point
• Diagnostic information for result validation

SQL Scripts

• Complete MLE workflows ready for execution
• Parameterized queries optimized for your data structure
• Result tables with proper schema for downstream analysis

Time Series Analysis Use Cases Supported

1. Structural break detection: Identify regime changes in time series
2. Change point analysis: Detect shifts in statistical properties
3. Regime changes: Find transitions between different data-generating processes
4. Anomaly detection: Locate abrupt changes in sensor or financial data

Key Parameters for ChangePointDetection

• TargetColumn: The numeric column to analyze for change points
• SegmentationMethod: Statistical model for segments (e.g., 'normal_distribution')
• SearchMethod: Algorithm for finding change points ('binary' or 'linear')
• MaxChangeNum: Maximum number of change points to detect
• Penalty: Information criterion for model selection ('BIC', 'AIC', etc.)
• OutputOption: Type of output ('changepoint', 'segment', 'verbose')

MLE Best Practices Applied

• PARTITION BY optimization for parallel processing across series
• ORDER BY with proper temporal column for time series ordering
• Parameter tuning specific to ChangePointDetection
• Result persistence using CREATE TABLE AS patterns
• Error handling for MLE-specific scenarios
• Scalability considerations for production workloads

Example Usage

-- Example ChangePointDetection workflow
-- Replace parameters with your specific requirements

-- 1. Execute ChangePointDetection
SELECT * FROM ChangePointDetection (
    ON your_database.your_timeseries_table AS InputTable
    PARTITION BY series_id
    ORDER BY timestamp_col
    USING
    TargetColumn('value_col')
    SegmentationMethod('normal_distribution')
    SearchMethod('binary')
    MaxChangeNum(10)
    Penalty('BIC')
    OutputOption('changepoint')
) AS dt;

Scripts Included

Core MLE Scripts

• mle_data_preparation.sql: Data preparation for MLE processing
• td_change_point_workflow.sql: Complete ChangePointDetection implementation
• table_analysis.sql: Time series structure analysis
• parameter_optimization.sql: Function parameter tuning

Integration Scripts

• mle_pipeline_template.sql: Multi-step analytical workflows
• performance_monitoring.sql: Execution monitoring
• result_interpretation.sql: Output analysis and visualization

Industry Applications

Supported Domains

• Economic forecasting and financial analysis
• Sales forecasting and demand planning
• Manufacturing process monitoring and quality control
• IoT sensor data analysis and alerting
• Network monitoring and traffic analysis
• Energy load and consumption pattern analysis

Limitations and Considerations

• MLE licensing: Requires proper Teradata MLE licensing
• Data ordering: Time column must support meaningful ORDER BY
• Computational complexity: Large datasets with many partitions may be resource-intensive
• Data quality: Results depend on clean, well-structured time series data
• Parameter sensitivity: Function performance depends on proper parameter tuning
• Temporal consistency: Irregular sampling may affect detection quality

Quality Checks

Automated Validations

• Time series structure verification
• Partition key distribution checks
• Parameter validation for ChangePointDetection
• Result quality assessment

Manual Review Points

• Parameter selection appropriateness
• Result interpretation accuracy
• Performance optimization opportunities
• Integration with existing workflows

Updates and Maintenance

• MLE compatibility: Tested with latest Teradata Vantage releases
• Performance optimization: Regular MLE-specific optimizations
• Best practices: Updated with Teradata community recommendations
• Documentation: Maintained with latest MLE features

---

This skill provides production-ready time series change point detection using Teradata's Machine Learning Engine ChangePointDetection function with industry best practices for scalable analytics.