ML operations entities created in Atlas

Each ML operations entity in Atlas includes detailed metadata collected from MLX.

The following diagrams show a summary of the entities created in Atlas for ML operations and assets. The supertypes that contribute attributes to the entity types are shaded.

Figure 1. Atlas Entity Types for ML Assets


Figure 2. Atlas Entity Types for ML Processes


The metadata collected for each entity type is as follows:

ML Project

Identifier Description and example content
typeName ml_project
guid System generated ID. This value is used to identify the entity in the Atlas Dashboard URL.
qualifiedName

???

name Project name as reported from MLX.
description String description metadata from MLX.
owner Owner as reported from MLX.
createTime Project creation time as reported by MLX.
modifiedTime Time from MLX indicating a change to the project. Formatted as in this example: “Wed Apr 17 2019 18:32:14 GMT-0700 (Pacific Daylight Time)”
ownerType Reserved for future use.
replicatedFrom Reserved for future use.
replicatedTo Reserved for future use.
Relationship: builds One project to many ml_model_build entities. ml_project_model_build

ML Model Build

Identifier Description and example content
typeName ml_model_build
guid System generated ID. This value is used to identify the entity in the Atlas Dashboard URL.
qualifiedName

???

name Model build name as reported by MLX.
description String description metadata from MLX.
createTime Time from MLX indicating the creation time of the model build. Formatted as in this example: “Wed Apr 17 2019 18:32:14 GMT-0700 (Pacific Daylight Time)”
owner Owner as reported from MLX.
version Integer value as reported from MLX.
metadata Key-value pairs reported from MLX.
defaultCpuMillicores ???
defaultMemoryMb ???
defaultGpus ???
imageTag URL to the container image for the model build.
imageHash Image signature. Combined with imageTag to uniquely identify a model build
exampleRequest JSON string describing an example request to the model.
exampleResponse JSON string describing an example response to the model.
ownerTypeReserved for future use.
replicatedFromReserved for future use.
replicatedToReserved for future use.
Relationship: project One project to many ml_model_build entities. ml_project_model_build
Relationship: deployments One build to many deployments. ml_build_deployment

ML Model Deployment

Identifier Description and example content
typeName ml_model_deployment
guid System generated ID. This value is used to identify the entity in the Atlas Dashboard URL.
qualifiedName

???

name Model deployment name as reported by MLX.
description String description metadata from MLX.
createTime Time from MLX indicating the creation time of the model deployment. Formatted as in this example: “Wed Apr 17 2019 18:32:14 GMT-0700 (Pacific Daylight Time)”
owner Owner as reported from MLX.
deployedTime Time from MLX indicating the deployment time of the model deployment. Formatted as in this example: “Wed Apr 17 2019 18:32:14 GMT-0700 (Pacific Daylight Time)”
metadata Key-value pairs reported from MLX.
modelEndpointURL The URL to receive a request and send back the response (prediction).
ml_model_deployment_status Enumeration indicating where the deployment is in its lifecycle.
cpuMillicores ???
memoryMb ???
gpus ???
replicas ???
ownerTypeReserved for future use.
replicatedFromReserved for future use.
replicatedToReserved for future use.
Relationship: builds One build to many deployments. ml_build_deployment

ML Project Create Process

Identifier Description and example content
typeName ml_project_create_process
guid System generated ID. This value is used to identify the entity in the Atlas Dashboard URL.
qualifiedName

???

name ???
description String description metadata from MLX.
userName User who created the project as reported from MLX.
owner Reserved for future use.
ownerTypeReserved for future use.
replicatedFromReserved for future use.
replicatedToReserved for future use.
Relationship: input ???
Relationship: output One process to one project. output

ML Model Train Build Process

Identifier Description and example content
typeName ml_model_train_build_process
guid System generated ID. This value is used to identify the entity in the Atlas Dashboard URL.
qualifiedName

???

name ???
description String description metadata from MLX.
userName User who triggered the operation to train a model and create a model build.
owner Reserved for future use.
ownerTypeReserved for future use.
replicatedFromReserved for future use.
replicatedToReserved for future use.
Relationship: input One process to one project. input
Relationship: output One process to one model build. output

ML Model Deploy Process

Identifier Description and example content
typeName ml_model_deploy_process
guid System generated ID. This value is used to identify the entity in the Atlas Dashboard URL.
qualifiedName

???

name ???
description String description metadata from MLX.
userName User who triggered the operation to deploy a model build.
owner Reserved for future use.
ownerTypeReserved for future use.
replicatedFromReserved for future use.
replicatedToReserved for future use.
Relationship: input One process to one model build. input
Relationship: output One process to one model deployment. output