Skip to content

Latest commit

 

History

History
204 lines (158 loc) · 9.78 KB

CHANGELOG.md

File metadata and controls

204 lines (158 loc) · 9.78 KB

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Added

  • n/a

Changed

  • n/a

Deprecated

  • n/a

Removed

  • n/a

Fixed

  • n/a

Added

  • Add raster:bands required property name for describing mlm:input bands (see README - Bands and Statistics for details).
  • Add README warnings about new extension eo and raster versions.

Changed

  • Split ModelBands and AnyBandsRef definitions in the JSON schema to allow them to be referenced individually.
  • Move AnyBandsRef definition explicitly to STAC Item JSON schema, rather than implicitly inferred via mlm:input.
  • Modified the JSON schema to use a if check of the type (STAC Item or Collection) prior to validating further properties. This allows some validators (e.g. pystac) to better report the real error that causes the schema to fail, rather than reporting the first mismatching type case with a poor error description to debug the issue.

Deprecated

  • n/a

Removed

  • Removed $comment entries from the JSON schema that are considered as invalid by some parsers.
  • When mlm:input objects do NOT define band references (i.e.: bands: [] is used), the JSON schema will not fail if an Asset with the mlm:model role contains a band definition. This is to allow MLM model definitions to simultaneously use some inputs with bands reference names while others do not.

Fixed

  • Band checks against eo, raster or STAC Core 1.1 bands when a mlm:input references names in bands are now properly validated.
  • Fix the examples using raster:bands incorrectly defined in STAC Item properties. The correct use is for them to be defined under the STAC Asset using the mlm:model role.
  • Fix the EuroSAT ResNet pydantic example that incorrectly referenced some bands in its mlm:input definition without providing any definition of those bands. The eo:bands properties have been added to the corresponding model Asset using the pystac.extensions.eo utilities.
  • Fix various STAC Asset definitions erroneously employing mlm:model role instead of the intended mlm:source_code.

Added

  • Add the missing JSON schema item_assets definition under a Collection to ensure compatibility with the Item Assets extension, as mentioned this specification.
  • Add ModelBand representation using name, format and expression properties to allow derived band references (fixes crim-ca/mlm-extension#7).

Changed

  • Adds a job to .github/workflows/publish.yaml to publish the stac-model package to PyPI.

Deprecated

  • n/a

Removed

  • Field mlm:name requirement to be unique. There is no way to guarantee this from a single Item's definition and their JSON schema validation. For uniqueness requirement, users should instead rely on the id property of the Item, which is ensured to be unique under the corresponding Collection, since it would not be retrievable otherwise (i.e.: collections/{collectionID}/items/{itemID}).

Fixed

  • Fix the validation strategy of the mlm:model role required by at least one Asset under a STAC Item. Although the role requirement was validated, the definition did not allow for other Assets without it to exist.
  • Correct stac-model version in code and publish matching release on PyPI.

Added

  • Add pattern for mlm:framework, needing at least one alphanumeric character, without leading or trailing non-alphanumeric characters.
  • Add examples/item_eo_and_raster_bands.json demonstrating the original use case represented by the previous examples/item_eo_bands.json contents.
  • Add a description field for mlm:input and mlm:output definitions.

Changed

  • Adjust scikit-learn and Hugging Face framework names to match the format employed by the official documentation.

Deprecated

  • n/a

Removed

  • Removed combination of mlm:input with bands: null that could never occur due to pre-requirement of type: array.

Fixed

  • Fix AnyBands definition and use in the JSON schema to better consider possible use cases with eo extension.
  • Fix examples/item_eo_bands.json that was incorrectly also using raster extension. This is not fundamentally wrong, but it did not allow to validate the eo extension use case properly, since the raster:bands reference caused a bypass for the mlm:input[*].bands to succeed validation.

Added

Changed

  • disk_size replaced by file:size (see Best Practices - File Extension)
  • memory_size under dlm:architecture moved directly under Item properties as mlm:memory_size
  • replaced all hardware/accelerator/runtime definitions into distinct mlm fields directly under the STAC Item properties (top-level, not nested) to allow better search support by STAC API.
  • reorganized dlm:architecture nested fields to exist at the top level of properties as mlm:name, mlm:summary and so on to provide STAC API search capabilities.
  • replaced normalization:mean, etc. with statistics from STAC 1.1 common metadata
  • added pydantic models for internal schema objects in stac_model package and published to PYPI
  • specified rel_type to be derived_from and specify how model item or collection json should be named
  • replaced all Enum Tasks names to use hyphens instead of spaces
  • replaced dlm:task by mlm:tasks using an array of value instead of a single one, allowing models to represent multiple tasks they support simultaneously or interchangeably depending on context
  • replace pre_processing_function and post_processing_function to use similar definitions to the Processing Extension - Expression Object such that more extended definitions of custom processors can be defined.
  • updated JSON schema to reflect changes of MLM fields

Deprecated

  • any dlm-prefixed field or property

Removed

Fixed

  • n/a

Added

  • Added example model architecture summary text.

Changed

  • Modified $id if the extension schema to refer to the expected location when eventually released (https://schemas.stacspec.org/v1.0.0-beta.3/extensions/dl-model/json-schema/schema.json).
  • Replaced dtype field by data_type to better align with the corresponding field of raster:bands.
  • Replaced nodata_value field by nodata to better align with the corresponding field of raster:bands.
  • Refactored schema to use distinct definitions and references instead of embedding all objects within dl-model properties.
  • Allow schema to contain other dlm:-prefixed elements using patternProperties and explicitly deny other additionalProperties.
  • Allow class_name_mapping to be directly provided as a mapping of index-based properties and class-name values.

Deprecated

  • Specifying class_name_mapping by array is deprecated. Direct mapping as an object of index to class name should be used. For backward compatibility, mapping as array and using nested objects with index and class_name properties is still permitted, although overly verbose compared to the direct mapping.

Removed

  • Field nodata_value.
  • Field dtype.

Fixed

  • Fixed references to other STAC extensions to use the official schema links on https://stac-extensions.github.io/.
  • Fixed examples to refer to local files.
  • Fixed formatting of tables and descriptions in README.

Added

  • Initial release of the extension description and schema.

Changed

  • n/a

Deprecated

  • n/a

Removed

  • n/a

Fixed

  • n/a