Integration Data Factory to score dynamically into the future with Machine Learning RRS feed

  • Question

  • Afternoon,

    I’m creating an Azure Machine Learning Predictive Web service which will consumed in cohort with Azure Data Factory and other azure services. I'm trying to create a setup data factory will continually score using my ML web service (batch execution) dynamically for the future. Meaning, If I already want to predict daily sales for the next two weeks, it will dynamically know which dates to pull to score.

    The model will have three inputs all coming from Azure SQL tables. Transaction Data which will transformed in inside the experiment to provide historical and lagging factors as features (moving averages, etc), Weather Data (updated daily with both historical observations and forecasts) as well as temporal data such as Economic Indicators.

    The transaction data contains information about a number of different product sales.

    Example of How transaction data is in SQL table: I'm unable to attach links, I can provide them.
    Example of After it has been transformed in the experiment: I'm unable to attach links, I can provide them.

    I'm then attempted to add another layer of complexity and have the model score different specific items within the dataset as azure machine learning does not support multi label regression (I'm trying to predict sales for each item_name).

    From my understanding, we will need to re-run the activity multiple times for different items using web parameters to change which item is under consideration. My best idea so far was to create a date table and have Azure Data Factory only pull using Functions the next two weeks of values.

    Thursday, July 20, 2017 3:20 AM