Using trained models in Power Query

As you already saw in Chapter 5, Importing Unhandled Data Objects, you used to pass objects that were the result of complex, time-consuming processing (including a ML model) in a serialized format specific to the language you were using. At that point, it was very easy to deserialize the file and get the model ready to use in Power Query to predict the target variable from new observations. However, it is important to know the dependencies needed by the scoring function (which takes the new observations as input and returns the predictions) because they are closely related to how the model was trained.

For this reason, we recommend the following:

IMPORTANT NOTE

If you need to use a serialized ML model provided by a third party, make sure that whoever developed it also provides you with a working scoring function in order to avoid unnecessary headaches when predicting target values for unknown observations.

If you think...