Xai feautures#9
Open
DanaDagher wants to merge 8 commits into
Open
Conversation
- shap_explain.py: KernelExplainer wrapper for RouteNet-Fermi GNN - Perturbs 10 scalar path features per flow while keeping graph intact - Selects 20 low-delay + 20 high-delay flows from constant_bitrate test set - 256 SHAP perturbations per flow, median background reference point - shap_plots/: 4 result plots - mean_shap_bar.png: global feature importance ranking - low_vs_high_shap.png: importance comparison across delay regimes - shap_summary_dot.png: beeswarm summary plot - shap_waterfall_worst.png: deep-dive on worst-case flow Key finding: traffic and packets dominate (>0.18s SHAP each); eq_lambda/avg_pkts_lambda matter more in high-delay regime (9x/5x ratio); timing params (exp_max_factor, pkts_lambda_on, avg_t_off, avg_t_on, ar_a, sigma) have zero importance for constant_bitrate -- candidates for masking in Step 5. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- step4_analysis.py: deep-dive on SHAP results - Feature importance table with % contribution - Direction analysis (positive/negative effect on delay) - Low vs high delay regime comparison - Masking recommendation: keep 4 / mask 6 features (60% reduction) - 2 new plots: scatter (SHAP vs feature value) + cumulative importance - presentation.md: full slide content for IMT+Sogeti meeting - 12 slides covering Steps 1-4 done + Steps 5-8 planned - Tables, findings, and plot placeholders Key finding: traffic(49.7%) + packets(47.0%) + avg_pkts_lambda(1.95%) + eq_lambda(1.34%) = 100% of importance. Bottom 6 timing features are zero for constant_bitrate -> safe to mask in Step 5. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Connected to TRAVEL project, WP3 Task 3.2, IMT contribution plan - Added Phase 1 (survey) -> Phase 2 (implementation) progression - Full 22-input mapping table (path/link/queue/graph tensors) - Justified why SHAP targets 10 scalar path features (not all 22) - Justified 40 flows (20 low+20 high) - computational budget + contrastive design - Justified 256 perturbations per flow (10x feature count rule) - Justified constant_bitrate as baseline (all timing params = 0, clean) - Justified keeping eq_lambda + avg_pkts_lambda despite small global importance - Added TRAVEL architecture diagram (Phase 1 -> Phase 2 -> Contributions) - Added Steps 5-8 detail and scientific question for retraining 14 slides total covering full internship arc Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.