Teradata HNSW Vector Search

Teradata HNSW Vector Search

• Minimum Version: Teradata 20.00 |

Core Capabilities

• Automated table structure analysis via DBC.ColumnsV
• Dynamic SQL generation for HNSW
• Complete workflow from data preparation to results interpretation
• Data quality validation and preprocessing guidance
• Parameter optimization and tuning

Key Parameters

• IDColumn: Unique vector identifier
• TargetColumns: Column(s) containing vector data
• M: Connections per layer (default 16)
• EfConstruction: Build-time search width (default 200)
• DistanceMetric: 'COSINE', 'EUCLIDEAN', 'INNERPRODUCT'
• TopK: Number of nearest neighbors to return (Predict)
• EfSearch: Search-time accuracy parameter (Predict)

Use Cases

1. Semantic similarity search
2. Recommendation systems
3. Image/document retrieval
4. RAG (Retrieval-Augmented Generation)
5. Duplicate detection

Example Usage

-- HNSW execution
SELECT * FROM HNSW (
    ON {USER_DATABASE}.{VECTOR_TABLE} AS InputTable
    USING
    IDColumn ('{ID_COLUMN}')
    TargetColumns ('{VECTOR_COLUMN}')
    M (16)                             -- Neighbors per layer
    EfConstruction (200)               -- Construction search size
    DistanceMetric ('COSINE')          -- 'COSINE','EUCLIDEAN','INNERPRODUCT'
) AS dt;

Scripts Included

Core Analytics Scripts

• table_analysis.sql: Automatic table structure discovery
• preprocessing.sql: Data preparation and feature engineering
• model_training.sql: HNSW execution
• evaluation.sql: Results analysis and metrics
• complete_workflow_template.sql: End-to-end workflow

Utility Scripts

• data_quality_checks.sql: Comprehensive data validation
• parameter_tuning.sql: Parameter optimization
• diagnostic_queries.sql: Results diagnostics and interpretation

Best Practices

• Always run table_analysis.sql first to understand your data structure
• Validate data quality before executing the analytical function
• Use parameter_tuning.sql to find optimal configuration
• Review diagnostic_queries.sql output for model/results validation

Limitations

• Requires Teradata Vantage 20.00+ with ClearScape Analytics
• Input data must meet function-specific requirements
• Results depend on data quality and parameter configuration

Teradata HNSW Vector Search - ClearScape Analytics skill for Teradata Vantage