Initial import: XAUUSD news sentiment models, evals, and data tooling

TF-IDF baseline (news_sentiment.py, models/xauusd_sentiment.joblib) and the
two-head DeBERTa-v3-small transformer (transformer/, news_transformer.py)
with its Google News crawler and DeepSeek labeller (data_collect/), the
golden-set and archive evals (eval/), and the 2026-09-22 and 2026-09-27
results and progress notes.

Transformer weights (model.pt, model.onnx) and the crawl database are not
tracked; rebuild with transformer/train.py and transformer/export_onnx.py.
The DeepSeek label cache in data/ is tracked.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This commit is contained in:
Harwinder Sekhon
2026-09-27 19:22:30 -04:00
commit 1d72e4006a
98 changed files with 1719757 additions and 0 deletions
+41
View File
@@ -0,0 +1,41 @@
precise: 200 windows of 14 days x 21 queries; 568 done, 4200 to go
100 requests, window 2024-07-15, 22643 unique headlines, 27 req/min
precise: 200 windows of 14 days x 21 queries; 729 done, 4039 to go
100 requests, window 2023-01-30, 23483 unique headlines, 27 req/min
200 requests, window 2026-08-24, 24041 unique headlines, 26 req/min
300 requests, window 2024-06-17, 24439 unique headlines, 26 req/min
400 requests, window 2020-01-06, 25024 unique headlines, 26 req/min
500 requests, window 2020-07-06, 25283 unique headlines, 26 req/min
600 requests, window 2023-03-13, 25737 unique headlines, 26 req/min
700 requests, window 2021-04-12, 26016 unique headlines, 26 req/min
800 requests, window 2022-10-24, 26529 unique headlines, 26 req/min
900 requests, window 2021-07-05, 27145 unique headlines, 26 req/min
1000 requests, window 2026-01-26, 27546 unique headlines, 26 req/min
1100 requests, window 2021-09-27, 28253 unique headlines, 26 req/min
1200 requests, window 2019-06-24, 28620 unique headlines, 26 req/min
1300 requests, window 2020-10-26, 28891 unique headlines, 26 req/min
1400 requests, window 2024-01-29, 29175 unique headlines, 26 req/min
1500 requests, window 2019-07-22, 29686 unique headlines, 26 req/min
1600 requests, window 2020-09-28, 30214 unique headlines, 26 req/min
1700 requests, window 2026-07-27, 30736 unique headlines, 26 req/min
1800 requests, window 2024-03-11, 31133 unique headlines, 26 req/min
1900 requests, window 2025-06-30, 31755 unique headlines, 26 req/min
2000 requests, window 2022-06-06, 32098 unique headlines, 26 req/min
2100 requests, window 2022-09-12, 32741 unique headlines, 26 req/min
2200 requests, window 2020-04-13, 33019 unique headlines, 26 req/min
2300 requests, window 2023-01-16, 33296 unique headlines, 26 req/min
2400 requests, window 2020-03-02, 33814 unique headlines, 26 req/min
2500 requests, window 2021-12-06, 34301 unique headlines, 26 req/min
2600 requests, window 2024-10-21, 34714 unique headlines, 26 req/min
2700 requests, window 2023-10-23, 35117 unique headlines, 26 req/min
2800 requests, window 2024-12-16, 35633 unique headlines, 26 req/min
2900 requests, window 2021-03-01, 35912 unique headlines, 26 req/min
3000 requests, window 2021-05-24, 36290 unique headlines, 26 req/min
3100 requests, window 2019-11-25, 36572 unique headlines, 26 req/min
3200 requests, window 2024-02-12, 37049 unique headlines, 26 req/min
3300 requests, window 2021-01-04, 37516 unique headlines, 26 req/min
3400 requests, window 2022-06-20, 37890 unique headlines, 26 req/min
3500 requests, window 2019-09-16, 38328 unique headlines, 26 req/min
precise: 200 windows of 14 days x 21 queries; 6125 done, 523 to go
100 requests, window 2024-03-25, 44823 unique headlines, 27 req/min
200 requests, window 2021-10-11, 45308 unique headlines, 27 req/min