4.2.5. Visualization

The Python library matplotlib was used to visualize the data in Python. This allowed an uncomplicated presentation of numerical data in 2D or 3D. Since the feature vector contained three variables (sputter power *P* sputter, gas flow *ϕ* and bias voltage *U*bias), a three-dimensional presentation of the feature space was particularly suitable. Here, the three variables were plotted on the x-, y- and z-axis and the measurement points were placed in this coordinate system. For the presentation of the corresponding numerical target value, color-coding serves as the fourth dimension. The target value of the measuring point could then be inferred from a color bar.

This presentation was especially suitable for small data sets, e.g., to get an overview of the actual position of the training data points. For large data sets with several thousand data points, a pure 3D visualization is too confusing, since measurement points inside the feature space were no longer visible. For this reason, a different visualization method was used to display the results of ML prediction of uniformly distributed data.

This visualization method is based on the visualization of computer tomography (CT) data set using a slice-based data view. Here, the 3D images of the body are skipped through layer-by-layer to gain insights into the interior of the workings level-by-level. Similar to this principle, the feature space was also traversed layer-by-layer.

Two feature variables span a 2D coordinate system. The measured values were again colored and displayed in the x–y plane analogous to the 3D display.

The third feature vector served as a run variable in the z-axis, i.e., into the plane. Employing a slider, the z-axis can be traversed, and the view of the feature space was then obtained layer-by-layer.
