ML Consuming Blueprints

What are ML Consuming Blueprints?

ML Consuming Blueprints are templates for a retrieving predictions from a model, that is served by a Model-Serving-Service. It showcases how to consume the model predictions in a microservice architecture via the FastIoT framework.

Pytorch Regression Consuming Blueprint

Pytorch Regression Consuming

import asyncio
import logging
import pprint
import random

from fastiot.core import FastIoTService, loop
from datetime import datetime

from blueprint_dev_v2.ml_lifecycle_utils.ml_lifecycle_broker_facade import request_get_prediction


class MlConsumerService(FastIoTService):

    @staticmethod
    def _get_random_raw_datapoint() -> dict:
        return {
            'laborant': ["TK", "HANS", "AN", "SO"][random.randint(0, 3)],
            'material_id': ["00000000", "11111111", "22222222", "33333333"][random.randint(0, 3)],
            'datum':  datetime.now().strftime("%d.%m.%Y, %H:%M:%S"),
            'rohwert_1_labormessung': random.uniform(0, 30),
            'rohwert_2_labormessung': random.uniform(0, 30),
            'rohwert_3_labormessung': random.uniform(0, 2),
            'aufbereiteter_wert''':  0.1,
        }

    @loop
    async def request_prediction(self):
        self._logger.info("Requesting prediction")
        raw_unlabeled_datapoints = [self._get_random_raw_datapoint() for _ in range(2)]
        self._logger.info(f"Requesting predictions for: \n{pprint.pformat(raw_unlabeled_datapoints)}")
        predictions = await request_get_prediction(fiot_service=self, data=raw_unlabeled_datapoints)
        self._logger.info(f"Received predictions: \n{pprint.pformat(predictions)}")
        return asyncio.sleep(5)


if __name__ == '__main__':
    # Change this to reduce verbosity or remove completely to use `FASTIOT_LOG_LEVEL` environment variable to configure
    # logging.
    logging.basicConfig(level=logging.DEBUG)
    MlConsumerService.main()