Visualization

DynaSchedBench visualization commands produce schedule plots, metric summaries, metric curves, event timelines, and an interactive Gantt inspector.

Machine Gantt

dsbx-vis gantt \
  -t runs/minimal/spt/trajectory_light.jsonl \
  -o runs/minimal/spt/gantt.pdf

Useful options:

• --label op: show operation labels such as O_{i,j}.

• --label job_id: show raw job IDs.

• --label job_op: show job ID plus operation index.

• --chunk 100: split long schedules into a multi-page PDF.

• --x-grid-step 20: fix vertical grid spacing.

• --warm --warmup-ratio 0.3: plot only the trailing warm-window.

Job-Centric Gantt

dsbx-vis job-gantt \
  -t runs/minimal/spt/trajectory_light.jsonl \
  -o runs/minimal/spt/job_gantt.pdf

This view places jobs on the y-axis and colors operations by machine group. It is useful for inspecting flow time, waiting time, and route structure.

Metrics Summary

dsbx-vis metrics-summary \
  -t runs/minimal/spt/trajectory_light.jsonl \
  -o runs/minimal/spt/metrics_summary.pdf

This recomputes scalar metrics and plots grouped diagnostic panels.

Metric Curve

List supported curve names:

dsbx-vis metrics-list

Plot one metric over time:

dsbx-vis metric-curve \
  -t runs/minimal/spt/trajectory_light.jsonl \
  -m wip \
  -o runs/minimal/spt/wip.pdf

Frequently useful curve names:

• wip

• wip_waiting

• wip_processing

• utilization_global

• queue_length_total

• jobs_completed

• stability_changed_ops_ratio

• reschedules_cumulative

Event Timeline

dsbx-vis events-timeline \
  -e runs/minimal/events.jsonl \
  -o runs/minimal/events_timeline.png

Without -o, this command opens an interactive matplotlib window. In headless or server environments, always pass -o.

Interactive Gantt Inspector

dsbx-vis gantt-inspect \
  -t runs/minimal/spt/trajectory_light.jsonl \
  -s runs/minimal/static_jobs.json \
  -e runs/minimal/events.jsonl

The inspector shows operation timing, planned processing times, and nearby events for the same job or machine when you hover over a segment.

Practical Tips

• Use trajectory_light.jsonl for large runs.

• Use --chunk when the Gantt chart is too dense for one page.

• Use --warm when comparing warm-start or long-horizon experiments.

• Save plots under the agent output directory so metrics, trajectories, and figures stay together.