FastRP (Fast Random Projection)
Generate node embeddings for machine learning
FastRP creates vector representations of nodes that preserve graph structure - similar nodes get similar vectors.
What It Computes
Fixed-size vector for each node (e.g., 128 dimensions).
When to Use It
- ML features: Use graph structure in ML models
- Similarity search: Find similar nodes efficiently
- Clustering: Cluster nodes in vector space
- Dimensionality reduction: Compress graph structure
Parameters
| Parameter | Type | Default | Description |
|---|---|---|---|
embedding_dim | int | 128 | Vector dimension |
normalization_strength | float | 0 | L2 normalization |
iteration_weights | list | [0, 1, 1] | multi-hop weights |
Performance
Time: O(E × dim)
Very fast compared to Node2Vec
Scales to: 100M+ edges
Example
Use Cases
Churn Prediction
Recommendation Systems
Find users similar to target user via embedding similarity
Fraud Detection
Cluster embeddings to find fraud rings
Link Prediction
Predict edges between nodes with similar embeddings
FastRP vs Node2Vec
| Aspect | FastRP | Node2Vec |
|---|---|---|
| Speed | Very fast | Slow (random walks) |
| Quality | Good | Better |
| Best for | Large graphs, speed | Quality critical |
FastRP is Raphtory's available embedding - Node2Vec not implemented
Temporal Embeddings
Generate embeddings for different time windows: