Trading Strategies
Optimize parameters for better returns and risk management.
experiment-3084112 / 240 trials
Start Free · 200 compute minutesDocker-native optimization
Your algorithm hasa better configuration.Let’s find it.
Bring your Docker image. Choose the parameters and the metric that matters. We run the trials and help you find what works best.
See workflow
How it works
You bring the question.We run the experiment.
Give your container a range of parameter values and a metric to improve. HyperOptimizer runs the trials and turns the results into a clearer next step.
Your Container
$ docker run strategy:latest \
--stop-loss=0.02 \
--take-profit=0.08 \
--lookback=120
Your image. Your parameters.
HyperOptimizer Platform
QueuedRunningFinishedNew bestFailed
Parallel trials, smart search, and automatic metric tracking.
Best Configuration
{
"stop_loss": 0.020,
"take_profit": 0.080,
"lookback": 120,
"score": 1580.4,
}
Waiting for improved trial results
Better results, faster.
Compare the strongest results and choose the trade-offs that matter to you.
One workflow. Many kinds of experiments.
ML Models
Tune hyperparameters to improve accuracy and reduce training time.
Simulations
Find the best scenario settings for complex simulations.
Data Pipelines
Optimize performance, cost, and reliability of your pipelines.
Custom Algorithms
Bring your own code and search space. We handle the rest.
PRODUCT
Every trial tells you something. See the whole picture.
Compare parameter choices, follow your objective metric, and inspect the logs behind each result. Keep the configurations worth exploring further.
Experiment dashboardBest result
Search space
learning_rate[0.001..0.01]batch_size[32,64,128]max_depth[3,5,8]
Trial dashboard
trial-108 running
trial-109 running
trial-110 queued
trial-104 best score: 0.942
Metric comparison
#configscorelat
1lr=0.003 b=640.942182
2lr=0.005 b=320.928171
3lr=0.001 b=1280.914194
Trial logs
$ python train.py --learning-rate=0.003 --batch-size=64
hpo.metrics.score=0.942
hpo.metrics.latency_ms=182
trial.status=completed
Experiment history
Revisit past experiments and compare their parameters and results within your plan’s history window.
WORKFLOW
How HyperOptimizer works
1. Package your workload
Bring a Docker image or containerized job.
2. Define the experiment
Set parameter ranges, objective metrics, trial count, and timeouts.
3. Run the search
We run the trials, collect their metrics, and use results to guide the next choices.
4. Compare outcomes
Rank configurations by your objective, then inspect their other metrics and logs.
FAQ
Frequently asked questions
Can't find the answer you're looking for? Reach out to oursupport team.
Is HyperOptimizer only for trading?
No. Trading strategies are one use case. HyperOptimizer is designed for any containerized workload with parameters, metrics, and an objective.
How is this different from Optuna, Ray Tune, or Katib?
Those tools give you optimization building blocks. HyperOptimizer gives you a place to run the whole experiment: launch containers, collect metrics, and compare results without setting up a cluster or results dashboard.
What do I need to use it?
A Docker image that accepts parameters and prints metrics. Choose which values to vary and which metric to improve; HyperOptimizer runs the trials.
Can I use custom metrics?
Yes. HyperOptimizer is metric-agnostic: optimize for accuracy, return, drawdown, latency, throughput, token cost, runtime, or any custom score you report.
Does my workload need to be machine learning?
No. Workloads can be ML models, trading strategies, simulations, data pipelines, LLM workflows, or any custom algorithm that can run in a container and emit metrics.
Can I run private workloads?
Yes. You can use private container images. HyperOptimizer runs the containers and processes their logs and metrics to show your results. Choose what your workload includes and what it writes to its output.
Do you support private Docker images?
Yes. Connect a supported private registry or use your workspace registry to make the image available to your trials.
How do compute limits work?
Free includes 200 compute minutes once per account. Starter and Pro include a monthly allowance, and you can buy prepaid top-ups. Choose a trial count and timeout for each experiment. If available compute runs low, execution stops; add compute and resume the experiment manually.
Do you support parallel optimization?
Yes. Free has 2 trial slots, Starter has 10, and Pro has 50. Running several trials at once lets you explore more configurations while the optimizer learns from completed results.
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.
Stop guessing configurations. Start running better experiments.
Your next experiment starts with the code you already have. Create an account, bring a Docker image, and put your first 200 compute minutes to work. Need help connecting your workload?Talk to us.