ibm/resource-yaml-creation

Creating ado Resource YAML Files

For CLI command syntax, see using-ado-cli. For full problem formulation workflow, see formulate-discovery-problem.

Metadata Fields

Every resource YAML should include a metadata block. The CLI uses these for display (ado get --details) and filtering (ado get --label, ado get --query).

metadata:
  name: my_space # short human-readable identifier
  description: | # longer explanation of purpose
    Optimize learning rate and batch size for ResNet training.
  labels:
    project: my_project # arbitrary key=value pairs for filtering
    team: ml_team

• name and description are shown by ado get --details
• labels support filtering: uv run ado get spaces --label project=my_project
• --query supports path-based filtering across any field: uv run ado get spaces --query "metadata.name=my_space"

Dynamic Reference Resolution

Resource YAMLs often reference other resources by ID. Leave these as placeholders and resolve them at creation time — do not hard-code IDs.

--use-latest

Queries the current context's metastore to find the most recently created resource of the given type.

# Create space, then operation that references it — no manual ID copy
uv run ado create space -f space.yaml
uv run ado create operation -f operation.yaml --use-latest space

--with

Creates a dependency inline and injects its ID automatically.

# Create space + actuatorconfiguration + operation in one command
# Note: You cannot use --with store=store.yaml or store_id here
# The space must use default store or have a valid store_id in the YAML
uv run ado create operation -f operation.yaml \
  --with space=space.yaml \
  --with actuatorconfiguration=config.yaml

Note: Can also specify resources ids to --with

uv run ado create operation -f operation.yaml \
  --with space=space-abcd-1234

--set

Overrides individual fields in the YAML at creation time without editing the file. Useful for environment-specific values or quick one-off changes.

# Override the sample store identifier
uv run ado create space -f space.yaml --set sampleStoreIdentifier=my_store

# Override a nested field using dot notation
uv run ado create operation -f operation.yaml --set parameters.budget=100

--set takes path=JSON_document pairs and can be used multiple times.

Validation

Always validate before creating:

uv run ado create RESOURCETYPE -f FILE --dry-run

--dry-run validates the YAML without creating the resource.

Templates

Use ado template to generate a starter YAML for any resource type:

# Generic discoveryspace template
uv run ado template discoveryspace

# Space template pre-filled for a specific experiment
uv run ado template discoveryspace --from-experiment my_experiment

# Operation template for a specific operator
uv run ado template operation --operator-name ray_tune

# ActuatorConfiguration template for a specific actuator
uv run ado template actuatorconfiguration --actuator-identifier my_actuator

Resource-Specific Guidance

DiscoverySpace

See formulate-discovery-problem for further details on creating discovery spaces.

Before creating (ado create space), check if a matching space already exists:

# Match by space config (entity space + experiments)
uv run ado get spaces --matching-space space.yaml

# Match by space ID
uv run ado get spaces --matching-space-id space-abc123

# Filter by label
uv run ado get spaces --label project=my_project --details

Reuse an existing space rather than creating a new one — it means the new operation benefits from measurements already collected.

Constitutive properties vs. parameterization:

• Declare a property as a constitutive property domain in the entity space when you want to explore a range of values for that property.
• Use experiment parameterization when you want to change the default value of an optional experiment property but keep it fixed across all entities. Do not add a single-valued domain to the entity space just to override a default.

# Correct: parameterization overrides experiment default, keeps it out of space
experiments:
  - actuatorIdentifier: trainer
    experimentIdentifier: train_model
    parameterization:
      - property:
          identifier: optimizer
        value: adam # overrides default "sgd"

# Incorrect: single-valued domain should be parameterization instead
entitySpace:
  - identifier: optimizer
    propertyDomain:
      variableType: DISCRETE_VARIABLE_TYPE
      values: [adam] # single value — use parameterization instead

Creating a space with a fresh samplestore:

uv run ado create space -f space.yaml --new-sample-store

ActuatorConfiguration

Before creating, check if a compatible configuration already exists:

uv run ado get actuatorconfigurations --details
uv run ado get actuatorconfigurations --label actuator=my_actuator

Reuse an existing actuator configuration when appropriate rather than creating duplicates.

SampleStore

You rarely need to create a samplestore explicitly. Every project comes with a default samplestore that is suitable for most use cases.

Create a new samplestore only when you explicitly want a clean slate with no shared measurement history.

uv run ado create samplestore -f samplestore.yaml

Related Resources

• using-ado-cli — CLI command syntax and shortcuts
• formulate-discovery-problem — full problem formulation workflow
• AGENTS.md — YAML testing and linting guidance