capo-machine-learning: typed, async Python SDK for AWS Machine Learning

capo-machine-learning is a typed Python SDK for AWS Machine Learning. It has an async client, AsyncMachineLearningClient, for asyncio and trio, and a sync client, MachineLearningClient, with the same methods. Requests and responses are typed with TypedDicts, so your editor and type checker know every field.

It is part of capo, a community-driven alternative to boto3 and aioboto3, built on its own HTTP stack instead of botocore and generated from the official AWS Smithy models. Documentation: https://capo-sdk.dev/services/machine-learning/

Installation

pip install capo-machine-learning

Usage

from capo_machine_learning import AsyncMachineLearningClient


async def main():
    async with AsyncMachineLearningClient() as machine_learning:
        # Example: call the add_tags operation
        response = await machine_learning.add_tags()
        print(response["resource_id"])

Pagination

Some operations in this SDK support pagination. If the operation supports pagination it will have an iter_ prefixed method that returns an async iterator.

from capo_machine_learning import AsyncMachineLearningClient


async def main():
    async with AsyncMachineLearningClient() as machine_learning:
        # Example: paginate over describe_batch_predictions
        async for item in machine_learning.iter_describe_batch_predictions():
            print(item)

Error Handling

The SDK raises exceptions for errors returned by the API. Catch them to handle failures gracefully.

from capo_machine_learning import AsyncMachineLearningClient
from capo_machine_learning.error import InternalServerException


async def main():
    async with AsyncMachineLearningClient() as machine_learning:
        try:
            await machine_learning.add_tags()
        except InternalServerException as e:
            print(f"Error: {e}")
            print(e.data)  # additional error data

Retrying

The SDK retries failed operations automatically. Retry behaviour follows the Smithy specification: errors are retried based on their is_retryable and is_throttling_error attributes. Throttling errors use a longer base delay. Network-level failures (connection errors and timeouts) are also retried. Non-retryable errors, such as client errors without the @retryable trait, are raised immediately without further attempts.

The number of attempts defaults to 3 and can be changed at the client level via retry_max_attempts, or per call via config_overrides.

from capo_machine_learning import AsyncMachineLearningClient


async def main():
    async with AsyncMachineLearningClient() as machine_learning:
        # Default: 3 attempts for every operation
        response = await machine_learning.add_tags()

        # Override per operation
        response = await machine_learning.add_tags(config_overrides={"retry_max_attempts": 5})

        # Disable retries for this call
        response = await machine_learning.add_tags(config_overrides={"retry_max_attempts": 1})