Curve Prediction

Description

Uses a trained model to predict and plot the response curve of an output against one of the inputs, while varying the other inputs with sliders and dropdown menus.

Application

One of the main applications of machine learning models is their speed at making new predictions. Once a model is trained on historical data, it can be used to make prediction (e.g. performance, strength, cost, …) instantaneously, compared to running the simulations and tests normally required to obtain these predictions. Plotting a curve rather than making a simple scalar prediction enables to get more insight and see how the output will vary against an input parameter.

How to use

You need a trained Model for Tabular Data to be able to use this step.

• Choose the Model you want to use to make the prediction.
• Choose the Output you want to predict. Unlike for the scalar prediction, you can only choose one output.
• If you have used the step Fix Parameters in this notebook, you could select a set of fixed parameters for that prediction.
• The y-axis will automatically vary to give the best reading of the curve. However, you can decide to Fix the range of the y-axis to some specific values. The range of the input sliders and of the x-axis is defined by the min and max values of the inputs in the training set of the model.
• Click Apply.
• You can then change the values of the inputs (either by changing the sliders in the step, or in a relevant step Fix Parameters), and the curve will be updated in a few seconds.

Examples

In the figure below, a model was used to predict the value of Output 1 based on seven inputs. The graph is showing the variation of the output against the variation of Input 6. Each time an input value is modified, the curve value changes accordingly. As the model used for the prediction was trained with the option to predict uncertainty, you can see a shaded range around the curve that indicates the uncertainty of the prediction. You can see that some predictions are more uncertain than others, and this can help you when making decisions.

In this example, you can also see that the uncertainty will vary with Input 6, and that the curve will sometimes be nearly linear, and some other times much more non-linear. Finally, in this example, the option to fix the range of the y-axis was not selected, and you can see the values changing for each prediction.