FOMC-RoBERTa stance scoring, point-in-time market pairing, dashboard filters - #1
Open
N041M wants to merge 1 commit into
Open
FOMC-RoBERTa stance scoring, point-in-time market pairing, dashboard filters#1N041M wants to merge 1 commit into
N041M wants to merge 1 commit into
Conversation
…filters NLP / stance scoring: - pipeline.py: fix FOMC-RoBERTa label order — resolve hawk/dove indices from the model config with a documented fallback (the published config ships generic LABEL_0/1/2; the hardcoded order was inverted, which would have flipped the sign of every stance score). Chunk long speeches and token-weighted aggregate, like analyze_documents, instead of scoring only the opening 512 tokens. - rescore_hawkish.py: snapshot the lexicon baseline before overwriting, backfill sentiment_signal.hawkish_score, and print a lexicon-vs-model comparison report. - analyze_hawkish_aggregation.py: report-only diagnostic comparing document aggregation schemes (mean / stance-weighted / drop-neutral / max-abs / net-label). Market pairing (replaces retired event_context): - features/market_response.py: relevance-aware, point-in-time pairing of each in-domain speech with the next-day move of the one market its speaker affects (no geography-blind 16x replication, no look-ahead). Pure helpers unit-tested. - analyze_habituation.py: share the abnormal-return computation (one source of truth). - features/sequence.py: carry sentiment/hawkish labels through the event sequence. - scheduler/jobs.py: stop building event_context (step9) on a schedule. Dashboard: - Migrate the Phase 1 Results tab off the stale event_context window-attribution onto the point-in-time pairing (fixes the spurious +/-1 strips). - Independent hide-neutral toggles (by sentiment and by stance) on the Phase 1 scatter, the Sentiment-over-time scatter, and the Speech Browser; surface the FOMC stance score/label columns in the browser. Tests: +tests/test_market_response.py, +FOMC label-resolver tests. 118 passing.
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 of this branch:
NLP / stance scoring
LABEL_0/1/2; indices are now resolved from the config with a documented fallback (label_0=dovish, label_1=hawkish). The old hardcoded order was backwards and would have flipped the sign of every stance score.analyze_documents) instead of only the opening 512 tokens.rescore_hawkish.py: snapshot the lexicon baseline, backfillsentiment_signal, print a lexicon-vs-model comparison.analyze_hawkish_aggregation.py: report-only diagnostic over document-aggregation schemes.Market pairing (replaces retired
event_context)features/market_response.py: relevance-aware, point-in-time pairing of each in-domain speech with the next-day move of the one market its speaker affects (no geography-blind 16x replication, no look-ahead).event_contextwindow-attribution (fixes the spurious ±1 strips).event_context.Dashboard
Tests: +
test_market_response.py, +FOMC label-resolver tests. ruff clean, 118 passing.