Kubeflow Katib: run a hyperparameter tuning Experiment with the Katib Python SDK (or Experiment YAML)

Steps

1. Install the Katib control plane and Katib Python SDK (`katib` package) per the installation guide
2. Define a Python objective function that computes and prints your target metric in `<metric-name>=<metric-value>` format, and a search space with `katib.search.int(min=, max=)`, `katib.search.double(...)`, or categorical values
3. Launch the Experiment via `KatibClient(namespace="kubeflow").tune(name=..., objective=objective, parameters=parameters, objective_metric_name="result", max_trial_count=12, resources_per_trial={"cpu": "2"})`
4. For non-Python or more complex training code, instead package training code in a container and configure an Experiment YAML with `objective` (type maximize/minimize, objectiveMetricName), `algorithm`, `parallelTrialCount`, `maxTrialCount`, `maxFailedTrialCount`, `parameters`, and `trialTemplate`, then `kubectl create -f experiment.yaml`
5. Wait for completion with `katib_client.wait_for_experiment_condition(name=name)` and fetch the winner with `katib_client.get_optimal_hyperparameters(name)` (or `kubectl get experiment <name> -o jsonpath='{.status.currentOptimalTrial}'`)
6. Review all Trials and their metrics in the Katib UI