Shift / Drift

Distribution-shift controls for graph, mechanism, and noise.

Use shift workflows when you want controlled distribution drift while preserving deterministic seeds and interpretable scale semantics.

When to use

  • You need train/test distribution shift for robustness evaluation.
  • You want independent control over graph, mechanism, and noise drift.
  • You want shift-aware observability in metadata and diagnostics coverage.

Shift modes

Mode-only examples:

shift:
  enabled: true
  mode: graph_drift

shift:
  enabled: true
  mode: mechanism_drift

shift:
  enabled: true
  mode: noise_drift

shift:
  enabled: true
  mode: mixed

Custom mode with explicit scales:

shift:
  enabled: true
  mode: custom
  graph_scale: 0.6
  mechanism_scale: 0.2
  variance_scale: 0.4

Scale interpretation

  • graph_scale: edge-odds multiplier is exp(ln(2) * graph_scale). Start at 0.5 for moderate structure drift.
  • mechanism_scale: increases probability mass on nonlinear mechanism families. Start at 0.5 for moderate mechanism tilt.
  • variance_scale: variance multiplier is exp(ln(2) * variance_scale). Start at 0.5 (+1.5 dB) for moderate noise drift.

Generation workflows

Run any shift-enabled config:

dagzoo generate --config path/to/shift_config.yaml --num-datasets 25 --out data/run_shift

Use discoverable smoke presets:

dagzoo generate --config configs/preset_shift_graph_drift_generate_smoke.yaml --num-datasets 25 --out data/run_shift_graph
dagzoo generate --config configs/preset_shift_mechanism_drift_generate_smoke.yaml --num-datasets 25 --out data/run_shift_mechanism
dagzoo generate --config configs/preset_shift_noise_drift_generate_smoke.yaml --num-datasets 25 --out data/run_shift_noise
dagzoo generate --config configs/preset_shift_mixed_generate_smoke.yaml --num-datasets 25 --out data/run_shift_mixed

What to inspect

  • Per-dataset metadata.ndjson records include resolved mode/scales and derived multipliers (edge_odds_multiplier, noise_variance_multiplier, mechanism_nonlinear_mass).
  • Diagnostics coverage summaries include shift observability metrics such as shift_graph_scale, shift_edge_odds_multiplier, shift_mechanism_nonlinear_mass, and shift_noise_variance_multiplier.

Benchmark runs can surface shift_guardrails in summaries.