results

Paper-plots

In this directory you find all plots used in the manuscript.

Organization

The directory contains the following sub-directories:

• Motivation:
  • pre-trained_dt_models/: contains a folder with the pre-trained DT models obtained with dt_train.py.
  • motivation-results/: contains one folder per experiments with SHAP values extracted directly from the two agents embb (HT in the paper) and urllc (LL in the paper); it also contains a subfolder processing-times/ which contains logs per experiment of the processing times. Both are instrumental to the final results of shap_explanations/.
  • shap_explanations/: contains two subfolder, one for the actual shap-values (i.e., Fig. 3 of the paper) and one for the processing-times (i.e., Fig. 4 of the paper).
• EXPLORA core part:
  • graphs-agents/: contains a simplified version of the graphs (no attributes displayed for the sake of readability - one per agent-traffic scenario, i.e., embb-trf1, embb-trf2, urllc-trf1, urllc-trf2).
  • plots-interpretation-agents/: contains the main results, i.e., the plots that connect the type of action with its impact on KPIs that are portrayed in Fig. 7 and Fig. 13 of the paper (one per agent-traffic scenario, i.e., embb-trf1, embb-trf2, urllc-trf1, urllc-trf2).
  • dts-interpretations/: contains the figures with the DTs that are produced on top of the synthesized explanations from the attributed graphs, i.e., Fig. 8 and Fig. 14 of the paper (one per agent-traffic scenario, i.e., embb-trf1, embb-trf2, urllc-trf1, urllc-trf2).
• Action steering:
  • action-steering/: contains the results obtained with action-steering-processing.py and action-steering-plotting.py:
    • data-for-cdfs/: contains the intermediate results generated by action-steering-processing.py that are input to action-steering-plotting.py to produce the plots.
    • plot-kpis/: contains the plots obtained with action-steering-plotting.py later elaborated with TikZ to produce the Fig. 9 and Fig. 10 of the paper, i.e., min_reward_vs_baseline/ for Fig. 9, and max_reward_vs_baseline/ and imp_bitrate_vs_baseline/ for Fig. 10.
    • distribution-actions-steered/: contains the results obtained results obtained with action-steering-processing.py on the distribution of actions that the IMPROVE module of EXPLORA suggest to replace.