telco_network_recovery: harden the template + refresh expected results#71
Merged
Conversation
Learnings propagated from the summit-demo refactor of the same chain. Script (telco_network_recovery.py): - reset_index(drop=True) on the train/val/test split DataFrames. The val split is an .iloc slice starting mid-frame, so model.data()'s df[col][0] type-inference lookup raised KeyError: 0 — a hard crash before Stage 1 fit on newer SDK releases. - MIP solve wrapped in a Gurobi→HiGHS fallback. A customer whose prescriptive engine isn't Gurobi-licensed previously hit a hard crash at the payoff stage; now it falls back to the bundled open-source HiGHS solver automatically. pyproject.toml: pin relationalai 1.2.2 -> 1.4.2. runbook.md doc accuracy: - "8 concepts" -> "8 source-data concepts" (the script defines more — TowerFailureScore, RestorePlan, GNN task tables). - Stage 2 response now separates advised-MODEL coverage (572/1500, 38.1%) from the at-risk label rate (597/1500, 39.8%) — two distinct metrics that were conflated. - Stage 8: RestorePlan.binding_constraint is a single String, and the chosen rows are marked via TowerUpgradeOption.is_selected_upgrade — corrected from the stale "list" + "SelectedUpgrade view" wording. Verified end-to-end against RAI 1.4.2.
…2 run The headline figures in the README and runbook were captured from an older run. Refreshed every expected-result number to the end-to-end verification run on RAI 1.4.2 (the version this template now pins): 142 critical-restore towers (was 166) · 36 selected (was 39) 207 Gbps restored (was 214) · $4,997,992 / $5M binding (was 4,999,671) 194 install-weeks (was 195) · 17 BRONZE / 15 SILVER / 4 GOLD GNN: failure_intensity median 2.92, 139/190 towers > 1.5 Added a stochasticity note to both the README expected-output block and the runbook intro: the equipment-failure GNN is stochastic, so the exact figures shift run to run while the structural outcome (all 5 regions, budget binding, ~200 Gbps, ~36 towers) reproduces. The README heading is now "Representative output (one run on RAI 1.4.2)" rather than implying seed-exact reproducibility.
|
The docs preview for this pull request has been deployed to Vercel!
|
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.
Summary
Hardens the
telco_network_recoverytemplate against two crash bugs, bumps the SDK pin, fixes doc accuracy, and refreshes every expected-result figure to a verified RAI 1.4.2 run. Learnings propagated from the internal summit-demo refactor of the same chain.Changes
Crash fixes (
telco_network_recovery.py)reset_index(drop=True)on the train/val/test split DataFrames. The val split is an.ilocslice starting mid-frame, somodel.data()'sdf[col][0]type-inference lookup raisedKeyError: 0— a hard crash before Stage 1 on current SDK releases.Dependency
pyproject.toml:relationalai1.2.2→1.4.2.Doc accuracy (
runbook.md)RestorePlan.binding_constraintis a single String and chosen rows are marked viaTowerUpgradeOption.is_selected_upgrade— corrected from the stale "list" + "SelectedUpgrade view" wording.Expected-results refresh (
README.md+runbook.md)Verification
Ran
telco_network_recovery.pyend-to-end on RAI 1.4.2 — pipeline completed all 4 stages, MIP reached OPTIMAL, RestorePlan materialized. NoKeyError, no traceback. The refreshed figures are this run's output.Test plan
python telco_network_recovery.pyon a fresh checkout to confirm it completes (figures will differ — GNN is stochastic — but structure holds).