Outputs and Plotting
Write fitted results and visualize algorithm performance
Explicit Exports
build() fits a model. It does not automatically write output files from the outputs flags. Call the required method:
from pathlib import Path
Path("output").mkdir(exist_ok=True)
model.save_to_csv("output")
model.save_graphs("output")
model.save_for_web("output")
model.save_to_mat("output")
model.save_zip("analysis.zip", "output")
Create the output directory first. save_zip archives the existing directory after the selected exports have been written.
The outputs.csv, outputs.png and outputs.web fields are preferences for callers and orchestration. They remain valid configuration fields even though built-in file creation is explicit.
CSV and Geometry
CSV output includes coordinates, processed data, algorithm-selection summaries and footprints. Two-dimensional rings retain holes and disconnected components. Three-dimensional exports include native mesh geometry and a versioned manifest.
Read each mesh manifest's index-base and schema declarations instead of assuming every export uses the same convention. Python's TRACE mesh and CLOISTER mesh schemas have different index conventions. Do not interchange their face tables without interpreting the manifest.
Plotting
import matplotlib.pyplot as plt
space.plot_sources()
space.plot_portfolio()
space.plot_good(0)
space.plot_footprint(0)
plt.show()
Algorithm arguments accept a retained algorithm name or a zero-based index. These helpers return matplotlib Axes and accept an existing axes where supported. Plots use 2D scatter/polygons or 3D axes/meshes according to the fitted projection. Source plotting requires source labels.
Persistence Versus Export
CSV, PNG, web and MAT files are presentation/interchange outputs. Native Model.save or an InstanceSpace checkpoint preserves the trained objects needed for continued Python use. A MAT export is not a portable MATLAB classifier archive.