Create an experiment
An experiment connects your Docker image to a search space and objective metric. Once configured, HyperOptimizer can schedule trials and rank the results.
Experiment fields
- Name
image- Type
- container image
- Description
The image HyperOptimizer starts for every trial.
- Name
command- Type
- optional
- Description
Override the image default command only if needed.
- Name
parameters- Type
- search space
- Description
Parameter names, ranges, and choices to pass as
--hpo-*.
- Name
objective- Type
- metric key
- Description
The metric to maximize or minimize.
- Name
parallelism- Type
- integer
- Description
How many trial containers can run at the same time.
Recommended first run
Start with a narrow search space and a small trial budget. The first goal is to validate the contract: arguments arrive, the workload runs, and metrics are collected.
- Use two or three parameters, not your entire strategy surface.
- Choose a metric that always prints for successful runs.
- Start with low parallelism until data and image access are confirmed.
After launch
Watch the first few trials. If they fail, read the trial logs and check the troubleshooting reference. If they succeed, widen the search space and increase the trial budget.