HuCE---Human-centered-Explanations

Usage and examples:

To create the different relations, run the file:

python concepts_kandinsky/concept_single_object.py
python concepts_kandinsky/concept_x_axis_relations.py
python concepts_kandinsky/concept_y_axis_relations.py
python concepts_kandinsky/concept_alignment_relations.py

Folders:

• concepts_kandinsky
  • all_possible_graphs_generator.py: Generates all possible graphs with 2, 3, 4, nodes with variable number of edges - once symmetric once non-symmetric
  • concept_single_object.py: Generates all possible graphs and Prolog output for the single object concept
  • concept_x_axis_relations.py: Generates all possible graphs and Prolog output for the x-axis relations concept
  • concept_y_axis_relations.py: Generates all possible graphs and Prolog output for the y-axis relations concept
  • concept_alignment_relations.py: Generates all possible graphs and Prolog output for the alignment relations concept (left of, right of, ...)
• data
  • input_data: Any data that can be used as input to further processing This folder contains some images of the symmetric Kandinsky Pattern "true", "false" and "counterfactual" subfolders have images for each case
  • output_data: All data produced by our code. It contains images of the generated graphs and Prolog output
• docs: Documentation items like figures
• graph_generators: Functions that deal with the generation of graphs with particular properties, such as fully-connected
• plots: Plotting functions
• test_cases: Test-cases for the generated Prolog output. Most of them are checks of the form "if an object is left of another" then by consistency the opposite relation "right of" incorporating those objects should be present in the generated Prolog file
• utils: Utility functions

Folders:

Kandinsky Patterns github repository: https://github.com/human-centered-ai-lab/app-kandinsky-pattern-generator

Contact:

• Bettina Finzel: bettina.finzel@uni-bamberg.de
• Anna Saranti: anna.saranti@boku.ac.at