9. Machine Learning — Frequently Asked Questions¶

9.1. Questions on the ML_Learn_Auto options¶

Does the normalize option normalize only the inputs, or both inputs and targets?¶

The normalize option normalizes both the inputs and the outputs.

What does “use target dependency” do?¶

It is used when the target columns provided are part of the curves — for example, the YY point columns of a curve.

How is the best model selected?¶

Grid search chooses the best parameters for each individual model, using the specified grid-search CV type and score type. The models — each with its best parameters — are then compared using cross-validation, and the best model is chosen with the selected CV option and CV score type.

Is “path dependency” considered?¶

In this case the curves are time histories and they are monotonic, so the dependency among the target columns is already accounted for as points along the time history.

How do I see the best parameters for the chosen model?¶

There is a hyperparameters output that gives the details. For a linear model there is also a coefficient output that gives the coefficient values.

My target YY columns are not equally spaced. Does the model predict the correct Y values, and does the saved mathmodel’s predicted curve show the correct X values?¶

The X values are not part of the input or the target in the ML_Learn_Auto worker, so the distribution of the X values is not a concern — the focus is on how the Y values are distributed at the same X value.

In the saved mathmodel’s predicted curve, the X values show the index of the YY columns. The Viz_AI worker can be given the XX columns to reconstruct the predicted curve using the provided X values.

I have several ML_Learn_Auto workers using the same inputs to predict different outputs, and I have to enter the same input values for each prediction. Is there a simpler way?¶

Add an ML_Predict worker after the mathmodel_save worker, and chain it with the dataset to predict on. Once this is set up, running the workflow produces the predictions from every model without entering the same inputs for each one.

9.2. Questions on the Reporter¶

How do I use the Reporter to generate reports?¶

Add a Reporter after the ML_Predict worker and use the ML_Predict worker’s Predictions output as the Reporter’s input dataset. Then choose Add Page and add the visualizations.

How do I export the report to PowerPoint?¶

After the Reporter runs, use the PPT Blob output — clicking it exports the PowerPoint slides.

How do I generate a report from different datasets?¶

There are two ways:

Add a Reporter on each branch to generate the reports, then consolidate them.
Add the different datasets to the same Reporter input datasets, and add visualizations using the different datasets as needed.