Artifact Animation
Add and play
Drag Artifact Animation from Widgets → Artifacts / Data onto the canvas. It is separate from Artifact Gallery. Place multiple animation widgets to compare different artifact indices or presentations; each instance keeps its own settings.
Load artifacts with the Artifacts toolbar button or
--artifact <directory>. For a local Journal, an adjacent artifacts/
directory is discovered when no explicit root has been selected. Animation uses
only the current Study's completed-trial view, not trials from other Studies in
the same Journal. Canvas selections do not filter the animation or its plots.
Wait for Study loading to finish before playback or export.
Artifact # chooses the existing zero-based entry index within each trial.
The index is chosen before checking image formats: index 1 does not mean the
second image after non-image entries have been removed. Only readable PNG, JPG,
and JPEG entries become frames. Original filename/MIME metadata identifies
extensionless Optuna artifact files. Missing, unreadable, CSV, GIF, and WEBP
entries are skipped. An empty-state message appears if no eligible frames
remain.
| Control | Operation |
|---|---|
| Play / Pause | Start or stop playback. Pausing keeps the current frame. |
| Previous / Next | Step through the available image frames. |
| Frame slider | Seek to an available frame; the label shows its Study trial number. |
| Display duration | Set the common live/GIF frame duration with a logarithmic slider or numeric input: 20–10000 ms, in 10 ms steps, default 330 ms. FPS is a calculated reference, not an editable setting. |
| Loop | Return to the first frame after the last frame, or stop at the end. |
Frame skip is common to widget preview and GIF export. 0 selects every
eligible image, 1 selects every second image, and so on: start with the first
image, select every N + 1th, and append the final eligible image once if
needed. The setting is bounded by the eligible image count. Play/Pause,
Previous/Next, the Frame slider, and Loop all use this selected sequence. The
selected count and estimated duration per pass include the appended endpoint.
There is no target-duration or automatic adjustment.
Changing skip keeps the current image if it remains selected; otherwise preview moves to the nearest selected image at or before it. Preparation is safely retargeted, and the new selection is used by export as well. Skipped images and artifactless trials still contribute observations to the synchronized plots.
Frames follow ascending Study-local trial numbers, not Journal-wide IDs or filenames. Numbers remain zero-based and are never renumbered: failed/pruned trials and trials without an image can leave gaps. Study changes, reloads, and artifact rescans rebuild the frame list and reset playback safely. Image/prefix preparation can delay advancement; the widget waits rather than showing stale data.
Images are prepared for the display size and scale, and reused across Animation widgets within a shared 256 MiB estimated retained image/texture/prefetch budget. This is not a limit on total application memory. Playback prefetches upcoming frames; pausing favors nearby frames and seeking prioritizes the new position. Evicted images may need to be read again. This cache is independent of Gallery. Use reload or artifact rescan to refresh files modified in place; silent file updates are not guaranteed to appear immediately.
Presentation
Expand Presentation to configure the content:
- Trial number overlays the actual Study trial number on the image.
- Objective values and Parameter values show selected names beside the
image. Use their Select … menus to choose the values. Missing or
non-finite values display
—. - Pareto 2D is available with at least two objectives. Choose the X and Y objectives; these axes are a projection of the full-objective analysis.
- Optimization History shows the selected history objective.
Images retain their original aspect ratio, with unused space rather than stretching or cropping. Optional charts appear beside the image, stacked when both are enabled. Narrow widgets place charts below the image. Disabled content does not reserve its section. Resize the widget to give selected values and charts enough room. Embedded charts have fixed axes, not the normal pan/zoom controls.
Reading the synchronized plots
At frame Trial t, plots contain only observed results whose Study trial number is ≤ t, including results from trials without artifacts. Future results are not drawn. The current trial is highlighted when it has valid data for that plot. Fixed Study-wide axis bounds keep the scale stable across frames; those bounds do not imply that future points are included.
Pareto ranks/front are recomputed for the prefix, not taken from the final Study's front. Dominance uses all objectives with their minimize/maximize directions, even if only two are displayed. A point can leave the front when a later trial dominates it, and seeking backward restores the earlier front. Non-finite or missing objectives exclude a trial from Pareto analysis. Only verified Feasible trials enter a constrained front; Infeasible and Feasibility unverified remain separate displayed categories.
History uses real Study trial numbers and finite values of the selected objective. It distinguishes the same feasibility categories. Its Best Value line is the cumulative objective-only minimum/maximum, including finite infeasible and unverified observations; it is not a verified-feasible best line.
Export GIF
Choose Export GIF and a destination. Export always starts at the first selected frame, regardless of the current pause/seek position. It exports the same chronological sequence selected by Frame skip for widget playback, including the first and final eligible images without duplicating the endpoint. Use Play or the Frame slider to preview that sequence before exporting.
Sampled frames still show all observations ≤ their trial number, including skipped images and artifactless trials. Display duration sets the exact GIF frame delay as well as live playback timing; Loop controls repetition.
During export, the initiating widget's playback/settings are paused. A protected fixed-size presentation captures a snapshot of the Study, artifact list, selected values, axes, theme, and presentation settings through the same content renderer. The GIF includes the image, enabled values, annotations, and charts—not transport controls, settings, surrounding canvas, or other panels. Different source-image dimensions do not change the output dimensions.
Progress and Cancel remain responsive; Escape also cancels. Once every frame is acknowledged, Finalizing GIF… briefly disables cancellation while publishing the complete file. Playback/settings return to their previous state after completion, cancellation, or error. Encoding runs in the background, one frame at a time. The destination is replaced only after all frames are successfully captured and encoded. If a needed image cannot be prepared, export aborts rather than silently changing the snapshot sequence. Already cached images can remain available even if their source file subsequently changes.
Keep the window dimensions and display scale unchanged during capture. If the content cannot fit in the protected capture area, resize the widget or zoom out and retry. Capture failures, timeouts, and write errors are reported in the widget and a GIF export error dialog.
GIF is palette-limited, so colors may be quantized. Frame delays have GIF's 10-millisecond precision; the selected duration uses the same 10 ms steps. PNG/JPEG artifacts are the only inputs; GIF/WEBP animation and MP4/video export are not supported.