Entropy/IP

Entropy/IP models each hexadecimal nybble of the IPv6 address separately, grouping contiguous nybbles into segments whose entropy behaves similarly and then learning a Bayesian network across those segments. During generation the network samples a state for every segment and reassembles a synthetic address that preserves the statistical relationships seen in the seeds.

• Reference: Placeholder citation (coming soon).

Train

rmap train --seeds seeds/active.txt --output models/entropy-ip.bin entropy-ip \
  --max-parents 3

Generate

rmap generate --model models/entropy-ip.bin --count 5000 --unique \
  --output entropy-ip.txt

Configuration

• --max-parents <N> – cap the number of parent segments when building the Bayesian network (default 3). Larger values capture more dependencies but increase training time.

Model notes

• Serialized models contain the learned segment definitions and conditional probability tables. The generator is deterministic given its RNG seed, so pass a different --seed to rmap generate if you need multiple disjoint batches.
• Training requires at least a handful of seed addresses per segment; when entropy is low the algorithm collapses the segment into a single fixed value.