Detect channel trajectories by global optimal-path routing

An alternative to sniff_trajectory_braid(). Instead of flow's local mutual-dominance trunk edges, each node is routed to the successor on its globally cheapest path to a final-year node, where edge cost is -log(weight) (the Jaccard). The minimum-cost path is the maximum-product Jaccard path (highest geometric-mean coherence). Final-year nodes root one in-tree each (the watershed of a final community); the cheapest birth -> final chain is the central trajectory (tr::<group>), the rest are absorbed tributaries (tr1 … trN), mirroring the flow object exactly. "Flow" names the object kind, not the algorithm, so the result is a birddog_flow and passes validate_flow().

Usage

sniff_trajectory_channel(x, min_group_size = 10, jaccard_min = 0.05, k_out = 2)

Arguments

x

A sniff_trajectory_dag() object, a sniff_groups_lineage() object, or a docs_per_group tibble (the DAG is built internally).

min_group_size, jaccard_min, k_out

Passed to sniff_trajectory_dag() when x is not already a DAG object.

Value

A birddog_flow object (see sniff_trajectory_braid()) with the extra latest_departure column.

Details

Adds one column to trajectories: latest_departure, the year of the most recent birth that can reach the trajectory's tail by any time-respecting path (Wu et al. 2014 latest-departure). For a central this is the freshest input anywhere in the final community's ancestry; combined with start it separates old-but-freshly-fed communities from genuinely young ones.

References

Wu H, Cheng J, Huang S, Ke Y, Lu Y, Xu Y (2014). Path Problems in Temporal Graphs. Proc. VLDB Endowment 7(9).

See also

sniff_trajectory_braid(), validate_flow(), sniff_trajectory_dag()

Other trajectory detection: is_flow(), sniff_trajectory_braid(), sniff_trajectory_dag(), subset.birddog_flow(), validate_flow()