Skip to content

Fusion State

Status: historical

The old fusion cache belonged to the windowed GAT/VGAE training stack. That stack was removed when TemporalData became the primary training surface.

The current live model families consume temporal event batches directly:

  • temporal_event_classifier
  • temporal_rnn_classifier
  • temporal_gat
  • temporal_vgae

Future fusion work should start from temporal event predictions or temporal state features, then aggregate to segment/source-level decisions where needed. Do not build new code against the old fixed-width graph-level fusion state.