# Teradata HNSW Vector Search

> Approximate nearest neighbor search using HNSW algorithm

- Skill: `teradata-labs/teradata-hnsw-vector-search` (Agent Skill, multi-file: 11 files)
- Install (CLI): `npx skillmds@latest add teradata-labs/teradata-hnsw-vector-search`
- Raw SKILL.md: https://api.skillmd.com/api/skills/teradata-labs/teradata-hnsw-vector-search/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Data & Analytics
- Author: teradata-labs (https://skillmd.com/u/teradata-labs)
- Updated: 2026-09-10
- Page: https://skillmd.com/skills/teradata-labs/teradata-hnsw-vector-search

---


# Teradata HNSW Vector Search

| Property | Value |
|----------|-------|
| **Skill Name** | Teradata HNSW Vector Search |
| **Description** | Approximate nearest neighbor search using HNSW algorithm |
| **Category** | Vector Search |
| **Primary Function** | HNSW |
| **Framework** | MLE
- **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

```sql
-- 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*