ML Analytics

ML Analytics Suite

Comprehensive machine learning algorithms for clustering, dimensionality reduction, outlier detection, and embedding quality assessment—all in SQL.

Clustering Algorithms

K-Means Clustering

Lloyd's K-Means with k-means++ initialization for finding customer segments, topic clusters, and data grouping.

-- CPU K-Means
SELECT cluster_kmeans(
  'customer_data',  -- table
  'features',       -- vector column
  5,                -- number of clusters
  100               -- max iterations
);

-- GPU K-Means (23x faster)
SELECT cluster_kmeans_gpu(
  'customer_data', 'features', 5, 100
);

-- Get cluster assignments
SELECT id, cluster_id, centroid_distance
FROM neurondb_cluster_assignments('customer_data', 'features', 5)
ORDER BY cluster_id, centroid_distance
LIMIT 100;

O(n·k·i·d)

Time Complexity

23x GPU

Speedup on GPU

k-means++

Initialization

DBSCAN (Density-Based)

Density-based clustering that automatically discovers the number of clusters and identifies outliers.

-- DBSCAN clustering (auto-discovers cluster count)
SELECT cluster_dbscan(
  'customer_data',
  'features',
  0.5,      -- epsilon (neighborhood radius)
  5         -- min_points (minimum cluster size)
);

-- Get clusters and outliers
SELECT cluster_id, COUNT(*) as size
FROM neurondb_dbscan_assignments('customer_data', 'features', 0.5, 5)
GROUP BY cluster_id
ORDER BY cluster_id;

-- cluster_id = -1 means outlier

Dimensionality Reduction

PCA (Principal Component Analysis)

Reduce high-dimensional vectors to lower dimensions while preserving variance.

-- Reduce dimensions: 768 → 128
SELECT reduce_dimensionality_pca(
  'embeddings_table',
  'vector_column',
  128  -- target dimensions
);

-- Returns: {"components": 128, 
--           "explained_variance": [0.45, 0.23, 0.12, ...],
--           "total_variance_explained": 0.80}

-- 80% of information retained with 83% size reduction

Outlier Detection

Isolation Forest

Detect anomalies and unusual patterns in your vector data using Isolation Forest algorithm.

-- Detect outliers with 95% confidence
SELECT detect_outliers(
  'customer_data',
  'features',
  0.95  -- confidence level
) AS outlier_count;

-- Get outlier details
SELECT id, anomaly_score
FROM neurondb_outlier_scores('customer_data', 'features', 0.95)
WHERE is_outlier = true
ORDER BY anomaly_score DESC;