Skip to content

Parameter search, run in the cloud

Explore more settings.Find what works.

Give HyperOptimizer your Docker image, parameter ranges, and a metric. We test configurations in parallel in the cloud and rank the results.

Start 7-day trial See how it works

7-day Starter trial · payment details required

For ML models, trading strategies, and simulations.

Explore the parameter space

Illustrative trading experiment

120 trials10 at a time

Final portfolio value across stop-loss and take-profit settings Each dot represents one of 120 illustrative backtests. The horizontal axis shows stop loss from 1 to 10 percent; the vertical axis shows take profit from 2 to 25 percent. Trial 81 is the best found: 3 percent stop loss,12 percent take profit, and a final portfolio value of$12,840. Background shading interpolates between trial results; it does not represent additional measured runs.

Best found · #081

$12,840

3% stop loss · 12% take profit

Illustrative backtest · $10,000 starting portfolio.
Shading between trial results is interpolated.

Every point is a run of your code.

From your code to clear answers

One Docker image.Many informed experiments.

You bring

  • Your Docker image

    Your code and dependencies

  • Settings to explore

    The values you want to try

  • A metric to improve

    What a good result looks like

We guide the search in our cloud.

Bayesian optimization uses results to suggest the next settings.

One Docker image is reused for three illustrative parallel runs, each with different parameter settings. Compare the measured results; run 3 has the best result in this example. Completed results guide the next parameter suggestions. Your code stays the same.

Search, cloud execution, and results — managed together.

Illustrative search · same image, different settings

See what your program needs

What you get back

Every result. The settings behind it.

Example: trading strategyIllustrative backtest

Compare configurations

Higher is better
Three highest final portfolio values in the illustrative experiment.
TrialStop lossTake profitFinal value
#081 Best found3%12%$12,840
#110 2%15%$12,703
#116 10%21%$12,695

120 trials · $10,000 starting portfolio.
Same historical data. Every result keeps its settings.

Inspect output from run #081
$ docker run trading-strategy:latest \
  --hpo-stop-loss-pct 3 \
  --hpo-take-profit-pct 12

hpo.metrics.final_portfolio_value=12840
trial.status=completed

Compare every run

Rank results by the metric you choose.

Keep the exact settings

See which parameters produced each result.

Inspect the details

Open the logs and revisit experiments within your plan’s history window.

Start 7-day trial

7-day Starter trial · payment details required

For models, strategies, and simulations.

ML models

Tune model settings against your validation metric.

Connect a Python model

Trading strategies

Compare backtests across strategy parameters.

Explore the trading workflow

Simulations

Search inputs against a measurable outcome.

Connect a simulation

Another workload? Connect any Docker image that accepts parameters and reports numeric metrics.

Fits the code you already use

Connect your code. Start your search.

Keep your language and dependencies. No HyperOptimizer SDK required.

Read the setup guide
  1. Package it in a Docker image
  2. Accept settings as arguments
  3. Print the metric you want to improve

One command. Ten trials at once.

Define your search in hpo.toml, then run it from your terminal. We handle the cloud execution and send results to your dashboard.

Terminal
120 trials completeIn the cloud · 10 at a time

Recent trial results

  1. [..] Trial 117 succeeded — final_portfolio_value: 12002
  2. [..] Trial 118 succeeded — final_portfolio_value: 12139
  3. [..] Trial 119 succeeded — final_portfolio_value: 12276
  4. [..] Trial 120 succeeded — final_portfolio_value: 12413
Best of 120 trials$12,840 final value

Trial #081 · Stop loss 3% · Take profit 12%

hpo.tomlExcerpt
algorithm = "bayesian"
maxTrialCount = 120
parallelTrialCount = 10

[[parameters]]
name = "stop_loss_pct"
type = "int"
min = 1
max = 10

[[parameters]]
name = "take_profit_pct"
type = "int"
min = 2
max = 25

[[objectives]]
name = "final_portfolio_value"
type = "maximize"

Same illustrative backtest · simulated progress · time compressed

Experiment settings & search methods CLI guide

Trial & pricing

Start your 7-day trial.

Try Starter with your own code. Payment details are required. Runs use the compute included in your plan.

Start 7-day trial Compare plans & billing terms
Starter €29/month
10 concurrent trials · 120,000 compute seconds per month
Pro €99/month
50 concurrent trials · 600,000 compute seconds per month

FAQ

Before your first experiment.

Can't find the answer you're looking for? Reach out to our support team.

Will my Docker image work?

Your program needs to accept parameter values as command-line arguments, run once, and print a numeric metric. Keep your language, libraries, and dependencies. No HyperOptimizer SDK is required.

Check the Docker requirements
Can I run private workloads?

Yes. Publish your Docker image to your private workspace registry. HyperOptimizer runs the containers and collects their logs and metrics in your dashboard. External registries are available on Enterprise.

How do compute limits work?

Starter includes 120,000 compute seconds per month; Pro includes 600,000 compute seconds per month. Set a trial count and timeout for each experiment. Each run uses your plan’s included compute.

See compute, PAYG, and renewal terms
What happens if a trial fails or times out?

You can inspect its status and logs in the dashboard. Other trials can continue within your experiment settings and available compute.

Bring your code.
Find better settings.

Define your experiment. Run it in our cloud. Compare the results. Need a hand getting started? Talk to us.

Start 7-day trial Review the setup guide

7-day Starter trial · payment details required