missing GEO type

[bodcsoft]

I have been looking at using mlcroissant to see if we can make our data holdings friendly to AI researchers
we are starting to host more of our data within ERDDAP, which allows for access in multiple different formats.

working my way through building a croissant file, I noticed there doesn't seem to be a GEO datatype in the DataType class.

to get my example working, I made a few changes to the mlcroissant locally and thought I would mention them here.

I added a new line to

croissant/python/mlcroissant/mlcroissant/_src/core/constants.py

Line 224 in 3c3a7fa

class DataType:

so it now contains a GEO

class DataType:
    """Data types supported by Croissant."""

    AUDIO_OBJECT = namespace.SDO.AudioObject
    BOOL = namespace.SDO.Boolean
    BOUNDING_BOX = ML_COMMONS_V_1_0.BoundingBox
    DATE = namespace.SDO.Date
    FLOAT = namespace.SDO.Float
    IMAGE_OBJECT = namespace.SDO.ImageObject
    INTEGER = namespace.SDO.Integer
    SPLIT = ML_COMMONS_V_1_0.Split
    TEXT = namespace.SDO.Text
    URL = namespace.SDO.URL
->    GEO = namespace.SDO.GeoCoordinates

Then in

croissant/python/mlcroissant/mlcroissant/_src/core/data_types.py

Line 24 in 3c3a7fa

EXPECTED_DATA_TYPES: dict[term.URIRef, type] = {

set the dict as such

EXPECTED_DATA_TYPES: dict[term.URIRef, type] = {
    DataType.BOOL: bool,
    DataType.DATE: pd.Timestamp,
    DataType.FLOAT: float,
    DataType.INTEGER: int,
    DataType.TEXT: bytes,
    DataType.URL: bytes,
->    DataType.GEO: float
}

This has worked for my example, and I would be happy to suggest the edits in a PR, I wouldn't know what the knock-on effects for the tests would be tho

Cheers
Tom

[thogar-computer]

Oops, It seems that my password manager was a little too eager with the auto sign and picked the totally wrong account.

[omshinde]

Thanks. We're developing a geo-extension for Croissant - GeoCroissant which will include geo-coordinates. See here