ML Extension
- Title: ML
- Identifier:
https://asterisk.coop/taco/spec/extensions/ml/v1.0.0/schema.json - Namespace:
ml - Scope: Collection
- Maturity: Pilot
- Owner: Asterisk Labs
This extension says what a model reads and what it predicts. The collection tasks names the task; the Labels and Instrument extensions describe the classes and bands of those files.
Collection fields
| Field | Type | Description |
|---|---|---|
ml:inputs |
list | Structure declarations or metadata fields a model reads |
ml:targets |
list | Structure declarations or metadata fields a model predicts; at least one |
"ml:inputs": ["s2.tif"],
"ml:targets": ["mask.tif"]
An entry with a : is a metadata field, such as scene:category for scene classification; any other entry is a declaration in taco:structure. A metadata field MUST occur at only one level. Every entry MUST exist in the contract, MUST NOT repeat within its list, and MUST NOT be both an input and a target.
Level fields
None.
Writer
from taco.extensions.collection.ml import ML
taco.Collection(..., tasks=["segmentation"], ml=ML(inputs=["s2.tif"], targets=["mask.tif"]))