Pydantic: What is the proper way to create a numpy ndarray field

[nicolls1]

Question

For bugs/questions:

• OS: Mac
• Python version import sys; print(sys.version): Python 3.6.8 :: Anaconda, Inc.
• Pydantic version import pydantic; print(pydantic.VERSION): 0.30

I would like to make a custom data type like done here: https://pydantic-docs.helpmanual.io/#custom-data-types but use numpy.ndarray. I have made validators so that the object can construct but I can't seem to figure out how to properly define the schema for this. Is this something I should be able to do and how?

import pydantic
import numpy

class NumpyNDArray(numpy.ndarray):
    @classmethod
    def __get_validators__(cls):
        yield cls.validate
    @classmethod
    def validate(cls, v: Any) -> str:
        # validate data...
        return v


class RemoteExample(pydantic.BaseModel):
    numpy_sample: NumpyNDArray = pydantic.Schema(None, alias='numpy_sample')


a = numpy.ndarray((1,2,3))
re = RemoteExample(numpy_sample=a)
print(re.schema())

Traceback (most recent call last):
  File "test_resources.py", line 8, in <module>
    print(re.schema())
  File ".../python3.6/site-packages/pydantic/main.py", line 463, in schema
    s = model_schema(cls, by_alias=by_alias)
  File ".../python3.6/site-packages/pydantic/schema.py", line 235, in model_schema
    model, by_alias=by_alias, model_name_map=model_name_map, ref_prefix=ref_prefix
  File ".../python3.6/site-packages/pydantic/schema.py", line 553, in model_process_schema
    model, by_alias=by_alias, model_name_map=model_name_map, ref_prefix=ref_prefix, known_models=known_models
  File ".../python3.6/site-packages/pydantic/schema.py", line 581, in model_type_schema
    f, by_alias=by_alias, model_name_map=model_name_map, ref_prefix=ref_prefix, known_models=known_models
  File ".../python3.6/site-packages/pydantic/schema.py", line 294, in field_schema
    known_models=known_models or set(),
  File ".../python3.6/site-packages/pydantic/schema.py", line 524, in field_type_schema
    known_models=known_models,
  File ".../python3.6/site-packages/pydantic/schema.py", line 794, in field_singleton_schema
    raise ValueError(f'Value not declarable with JSON Schema, field: {field}')
ValueError: Value not declarable with JSON Schema, field: numpy_sample type=NumpyNDArray default=None

I tried looking through field_singleton_schema and couldn't find anything obvious to do.

[samuelcolvin]

see #380 for a similar discussion, apart from schema.

Regarding schema generation, I'm afraid it's not possible at the moment afaik.

I guess we should add support for a __get_schema__ or similar method on custom types. I thought we already had a dicussion about it, but can't find an issue.

@tiangolo thoughts?

[dgasmith]

+1 for __get_schema__, would you take a PR for this?

[samuelcolvin]

:+1:

[tiangolo]

Agreed. I've been planning on implementing/proposing __get_schema__ for custom types, but haven't found the time 😅

[dgasmith]

I haven't touched this yet and will be a few weeks on my end if you get to it first :)

[samuelcolvin]

This should be fixed by #1065. Please comment there if you have any feedback or suggestions.