6Tree

6Tree projects each seed address into vectors at several bases (binary through base32) and recursively splits the space on the dimension that best separates the remaining seeds. Leaves with enough support become address patterns that can be sampled to produce new targets.

Reference: Placeholder citation (coming soon).

Train

rmap train --seeds seeds/hitlist.txt --output models/6tree.bin six-tree \
  --base 16 --min-seeds-per-leaf 16

Generate

rmap generate --model models/6tree.bin --count 250000 \
  --unique --output 6tree.txt

Configuration

--base <2|4|8|16|32> – select the vector base for splitting (default 16).
--seed <u64> – optional RNG seed for deterministic generation (default 0).
--min-seeds-per-leaf <usize> – minimum number of seeds required to keep splitting (default 16).
--max-depth <usize> – optional cap on the tree depth for faster training.

Model notes

Different base projections expose distinct structural signals; consider building multiple models with varying bases and combining the generated targets.
The serialized model stores a tree specialized to the chosen base, so use the same base when regenerating documentation or comparing metrics.

6Graph DET