Open-source analysis toolkit

RGCAT

Roboforming Geometric Coverage Analysis Toolkit

A reproducible Python toolkit for cluster-based geometry coverage analysis, gap identification, and analytical method comparison in Roboforming process data.

Explore RGCAT Release status

This website previews RGCAT’s methods and outputs. Source code, documentation, and installation instructions will be published with a later public release.

Developed by Per Starke as part of a master’s thesis, supervised by Jan Schäfer.

Laptop showing an RGCAT analytics dashboard with run metrics, gap candidates, and reconstructed geometry plots

Overview

From geometry tables to reproducible analysis outputs.

RGCAT turns tabular Roboforming geometry data into reduced representations, clusters, coverage metrics, gap candidates, reconstructed candidate features, and timestamped run outputs.

Reproducible Runs

YAML configs, timestamped run folders, metrics, logs, plots, config snapshots, and run metadata.

Modular Analysis Pipeline

PCA, VAE, and VaDE for representation; KMeans, HDBSCAN, and combo modes for clustering; configurable gap analysis.

Extension Evaluation

Compare the original embedding with the same embedding plus selected analytical gap candidates.

Pipeline

A config-driven analysis flow.

Each run follows the same stage order; YAML controls representation, clustering, coverage metrics, and gap analysis.

  1. 01

    Data Ingestion

    CSV, Parquet, Feather

  2. 02

    Feature Expansion

    flat vectors, z-map features, context columns

  3. 03

    Representation

    PCA, VAE, VaDE

  4. 04

    Clustering

    KMeans, HDBSCAN, combo

  5. 05

    Coverage Metrics

    kNN coverage, sparse-region metrics

  6. 06

    Gap Detection

    candidate sampling, ranking

  7. 07

    Reconstruction

    candidate features, original space

  8. 08

    Extension Evaluation

    analytical before/after metrics

  9. 09

    Run Outputs

    metrics, plots, report, config snapshot

Outputs

Plots, metrics, reports, and config snapshots.

Each run writes a timestamped folder with logs, metrics, plots, config snapshots, run metadata, and an optional static report dashboard.

RGCAT kNN coverage map scatter plot

Coverage Map

k-nearest-neighbor distances highlight sparse regions in the embedding.

RGCAT embedding gap candidates plot

Gap Candidate Plot

Analytical gap candidates shown against existing embedding samples.

RGCAT reconstructed TCP z-map candidate heatmap and 3D surface

Reconstructed TCP Z-map

One analytical candidate reconstructed as heatmap and 3D surface.

RGCAT VAE reconstruction loss plot

VAE Reconstruction Loss

Training and validation loss for a VAE representation run.

Static RGCAT report.html dashboard with run KPIs, evaluation summary, gap candidate plot, and reconstructed geometry preview

Run Dashboard

A static reporting dashboard is created for each run.

When reporting is enabled in the config, every successful run creates runs/<run_id>/report.html. It opens in a browser without a server and summarizes the saved run artifacts.

  • KPIs and Evaluation Summary for the completed run.
  • Plot gallery for coverage, gap candidates, and reconstructed geometries.
  • Links to metrics.json, config.yaml, run_info.yaml, and logs.txt.

Configuration reference available with the public release.

Files created by a run

runs/<run_id>/logs.txt runs/<run_id>/metrics.json runs/<run_id>/config.yaml runs/<run_id>/plots/ runs/<run_id>/run_info.yaml runs/<run_id>/report.html

Release status

RGCAT is being prepared for public release.

The source repository is not yet publicly accessible. The software, reproducible environments, documentation, and license files will be published together in a later release.

Included with the public release

  • Python source code
  • Docker and Conda environments
  • Usage and configuration guides
  • Pipeline walkthrough
  • Architecture documentation and ADRs
  • MIT license

Method comparison

Compare pipeline variants with the same outputs.

Use the same data, folder structure, metrics, and analytical before/after evaluation to compare pipeline variants reproducibly.

Coverage before/after

Cluster changes

Reconstruction summary

Candidate diagnostics

Reconstructed candidates are analytical suggestions, not validated process data.

Documentation

Documentation included with the public release.

The release will include the complete usage, configuration, pipeline, and architecture reference set.

README
Usage Guide
Configuration Reference
Pipeline Walkthrough
arc42-based Architecture Docs
Architecture Decision Records

License

MIT License.

The license file will be published together with the source repository.

Read MIT License