Usage
Target Generation

Target Generation

Each algorithm exposed by rmap train and rmap generate is documented in this section. Every page explains how the technique works at a high level, links to the original research, highlights the configuration flags, and outlines what to expect from the resulting model.

Algorithms

  • Entropy/IP – Bayesian segmentation of high-entropy nybbles.
  • Random IP – Uniform baseline over the IPv6 address space.
  • 6Gen – Density-driven clustering of seed regions.
  • 6Forest – Ensemble decision forest over hierarchical partitions.
  • 6Graph – Pattern mining from nybble-level graphs.
  • 6Tree – Base-b divisive tree search over address vectors.
  • DET – Entropy-guided region expansion using space trees.
  • 6Probe – Address Space Forest construction with DHC variants.
  • 6GAN – Reinforcement-trained generative adversarial network.
  • 6GCVAE – Gated convolutional variational auto-encoder.
  • 6VecLM – Transformer language model over IPv6 tokens.

CLI Workflow

Training produces serialized model files that the CLI and Python bindings can re-use for generation or direct scanning.

# Train SixGen with a 1M address budget
rmap train --seeds seeds/observed.txt --output models/sixgen.bin six-gen --budget 1000000

# Produce 10k unique addresses from the trained model
rmap generate --model models/sixgen.bin --count 10000 --unique --output targets.txt

Common flags while you explore individual pages:

  • --seeds <FILE> – newline-separated IPv6 seed addresses.
  • --output <FILE> – destination path for the serialized model.
  • --model <FILE> – model file to load for generation.
  • --count <N> – number of addresses to emit, optionally guarded by --unique and --max-attempts.
  • --exclude <FILE> – drop addresses listed in the exclusion file before emission.

Generated datasets integrate directly with rmap scan --path and rmap analyze once you have a model that performs well for your target space.