Create a Pydantic model from JSON Schema dynamically at runtime

json_schema_to_pydantic_model.py

import tempfile
import json
from pathlib import Path
from typing import Any, Dict, Type

import datamodel_code_generator
from pydantic import BaseModel


def json_schema_to_pydantic_model(json_schema: Dict[str, Any]) -> Type[BaseModel]:
	"""Generate a Pydantic model from a JSON schema in memory."""

	# Create temporary files for input and output
	with tempfile.NamedTemporaryFile(
		mode="w+", suffix=".json", delete=False
	) as input_file, tempfile.NamedTemporaryFile(mode="w+", suffix=".py", delete=False) as output_file:
		# Write the JSON schema to the input file
		input_file.write(json.dumps(json_schema))
		input_file.flush()

		# Generate the model code
		datamodel_code_generator.generate(
			input_=Path(input_file.name),
			input_file_type=datamodel_code_generator.InputFileType.JsonSchema,
			output=Path(output_file.name),
			target_python_version=datamodel_code_generator.PythonVersion.PY_312,
			allow_population_by_field_name=True,  # causes json title's to be used as field aliases
		)

		# Read the generated code
		output_file.seek(0)
		generated_code = output_file.read()

	print(f"Generated code:\n\n{generated_code}")

	# Clean up temporary files
	Path(input_file.name).unlink()
	Path(output_file.name).unlink()

	# Create a namespace to execute the code in
	namespace = {}

	# Execute the generated code in the namespace
	exec(generated_code, namespace)

	# Convert the model name to a valid Python class name
	schema_root_item_name = json_schema["title"]
	model_name = datamodel_code_generator.parser.base.title_to_class_name(schema_root_item_name)
	model = namespace[model_name]
	assert issubclass(model, BaseModel)
	return model


if __name__ == "__main__":
	# todo: add a test case

	# Example usage
	json_schema = """
    {
        "title": "person",
        "type": "object",
        "properties": {
            "name": {"type": "string"},
            "age": {"type": "integer"},
            "pets": {
                "type": "array",
                "items": {
                    "type": "object",
                    "properties": {
                        "name": {"type": "string"},
                        "species": {"type": "string"}
                    },
                    "required": ["name", "species"],
                    "additionalProperties": true
                }
            }
        },
        "required": ["name", "age"],
        "additionalProperties": false
    }
    """

	PersonModel = json_schema_to_pydantic_model(
		json_schema=json.loads(json_schema),
	)

	print(f"Generated model: {PersonModel}")

	print(f"Model schema:\n\n{json.dumps(PersonModel.model_json_schema(), indent=2)}")