Compare two runs
See two submissions side by side. Parameters, best solutions, and convergence curves.
When you are tuning a genetic algorithm, the question is almost always the same: did that parameter change actually help? The comparison view answers it in one screen. Pick two of your past runs, and the page shows their parameters, best solutions, summary stats, and overlaid convergence curves side by side.
Starting a comparison¶
There are three places to kick off a comparison:
- From your submissions list. Open Your submissions list, tick the checkbox at the start of any two completed rows, and click Compare in the floating bar at the bottom of the page.
- From a submission's detail page. Click the Compare button next to the existing action buttons. The next page lets you pick which of your other runs to compare with.
- From a result page. Click Compare to compare the result you are currently viewing with another run. The result you came from stays selected as side A.
Only completed runs can be compared. Pending or failed runs are still visible on the list, but their checkboxes are disabled.
The comparison page¶
The page is divided into a header and three result sections.
Header¶
The top card tells you which problem you are looking at. If both runs came from the same problem definition, you see one title; otherwise both problem names appear so you can see at a glance that the runs are not directly comparable. Each side also shows its status badge, submission id, creation date, and a link to the full result.
If one of the runs has more than one result, a small dropdown next to its header lets you choose which result to compare against. The page reloads with your choice.
GA parameters¶
A table lists every comparable parameter (number of generations, population size, crossover and mutation settings, and so on) with the value used by each run. Rows where the two runs disagree are highlighted, so the actual differences jump out.
A Show differences only switch above the table hides identical rows when you only care about what changed.
Best solutions¶
Each side shows its best fitness value and the gene-by-gene best solution. When both runs share the same objective direction (both maximizing or both minimizing), the better fitness is highlighted in green and tagged with a trophy.
For multi-objective runs the highlight is intentionally omitted: trade-offs across a Pareto front are not strictly comparable, and a single "winner" would be misleading.
Convergence¶
A single chart overlays the two best-fitness-per-generation curves. Run A is drawn in the primary blue, Run B in the warning orange. Hover the chart to see both runs' fitness at the same generation. The chart respects your prefers-reduced-motion setting.
If one or both runs did not produce per-generation history (for example, an old run or one that crashed), the chart panel shows a clear empty state rather than failing.
Summary statistics¶
A small table compares four key numbers per run:
- Best fitness achieved.
- Generation when the best was first reached. Reaching the same fitness in fewer generations is usually a sign your settings converge faster.
- Total generations run.
- Improvement from the first generation to the last. Both the absolute change and the percentage relative to the starting fitness.
The better value in each row is highlighted. If a row cannot be compared (for example, a multi-objective run), the cell shows a dash instead.
Sharing a comparison¶
The comparison URL is reproducible. Bookmark it, paste it into a chat with a teammate, or save it for later. As long as you are signed in and both runs are yours, the same view loads.
Common questions¶
Why is the comparison "partial"? One or both runs are not fully complete (for example, still running, failed, or missing per-generation history). The page tells you which sections are affected and renders everything else normally.
Can I compare runs from different problems? Yes. The parameter diff and the convergence chart still work. Only the "Best solution" panel becomes less informative because the gene meanings differ between problems.
How do I swap one side without starting over? On the comparison page, click Swap one side. Pick which side to replace, then pick a new run. The other side stays as is.