Many of the systems we use support different kinds of metadata. To build up better tooling and improve workflows we should develop some metadata standards that allow us to map between the systems:
WandB
Argo/k8s
fv3net-artifacts/GCS
Logging/big-query
Name
Name
URL encoding date/project/step/tag
labels
Tags
Labels: key/value with limitations (searcheable)
jsonPayload
Groups
annotations (key/value unrestricted)
resource.labels
Config
Summary Metrics
Artifacts
I'm think for any artifact/workflow we will want to know the following
project
experiment
trial/run
dependencies
revision/commit
author
date created
user-facing configuration/yaml
We should develop some some standards for how to map these concepts onto our experiment management systems.
frodre
commented
2 years ago
Many of the systems we use support different kinds of metadata. To build up better tooling and improve workflows we should develop some metadata standards that allow us to map between the systems:
I'm think for any artifact/workflow we will want to know the following
We should develop some some standards for how to map these concepts onto our experiment management systems.