Machine learning, in motion
Build a model.
Take the track.
Swap the parts of a machine-learning pipeline.
Race the trade-offs. Refine your build.
Next on the grid
You’re in line.
Your build screen will open when a place is available. You can watch the current race below.
On track now
01 / The build
Choose your build.
About this challenge
Swapping ML libraries usually means rewiring data and evaluation. TEMPO gives each component a shared interface.
Your measured build
Recorded benchmarkCompute = extraction + selection + model fitting. Memory = peak RAM during extraction. Your choices replay recorded benchmarks; they do not run a new experiment.
Full component library
Only measured combinations are selectable. Choose a complete build to switch between components that have different compatible options.
How TEMPO makes this possible
A pipeline is a chain of plug-ins. TEMPO adapts datasets to one format, connects extractors, selectors and models, and evaluates them with the same experimental setup. Feature caching avoids repeating extraction; stage-level telemetry records time and memory. The framework handles that plumbing when a component changes.
Race rules: driving cost = 1 + 0.12 × ln(1 + compute seconds); accuracy pit cost = 4 × (1 − accuracy). Both costs use one shared scale so the final car finishes at the chosen race duration. This weighting is a game rule, not a benchmark measurement. Memory is compared separately.
This challenge uses one eligible benchmark measurement per build. The poster summarises separate experiments; this replay does not establish statistical significance.
On the grid
Choose a build, then join the grid.
02 / On track
Grand Prix.
- Elapsed
- 00:00.0
- Your position
- —
- Lap
- — / 2
Live positions
2 laps03 / The comparison
Across the line.
Technical details
Compare speed & predictions
Try changing one component and compare the result.