harry/News-ML-Decoder
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

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

Browse Source 
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
Download Patch File Download Diff File Expand all files Collapse all files
logs/label_full_20260927.log
+53
View File
@@ -0,0 +1,53 @@
DeepSeek deepseek-v4-flash, prompt d32e04c6; budget 200 calls
pools: precise 2778, broad 20532 unlabelled; picked 5000
selected 5000 unlabelled headlines
250 labelled, 10 calls (0 failed), 0 from cache
DeepSeek batch of 25 failed (attempt 1/3, None) — retrying
DeepSeek batch of 25 failed (attempt 2/3, None) — retrying
DeepSeek batch of 25 failed (attempt 3/3, None) — retrying
475 labelled, 22 calls (3 failed), 0 from cache
DeepSeek batch of 25 failed (attempt 1/3, None) — retrying
DeepSeek batch of 25 failed (attempt 2/3, None) — retrying
DeepSeek batch of 25 failed (attempt 3/3, None) — retrying
700 labelled, 34 calls (6 failed), 0 from cache
950 labelled, 44 calls (6 failed), 0 from cache
1200 labelled, 54 calls (6 failed), 0 from cache
DeepSeek batch of 25 failed (attempt 1/3, None) — retrying
DeepSeek batch of 25 failed (attempt 2/3, None) — retrying
DeepSeek batch of 25 failed (attempt 3/3, None) — retrying
1425 labelled, 66 calls (9 failed), 0 from cache
1675 labelled, 76 calls (9 failed), 0 from cache
1925 labelled, 86 calls (9 failed), 0 from cache
2175 labelled, 96 calls (9 failed), 0 from cache
2425 labelled, 106 calls (9 failed), 0 from cache
2675 labelled, 116 calls (9 failed), 0 from cache
2925 labelled, 126 calls (9 failed), 0 from cache
3175 labelled, 136 calls (9 failed), 0 from cache
DeepSeek batch of 25 failed (attempt 1/3, None) — retrying
3425 labelled, 147 calls (10 failed), 0 from cache
3675 labelled, 157 calls (10 failed), 0 from cache
3925 labelled, 167 calls (10 failed), 0 from cache
4175 labelled, 177 calls (10 failed), 0 from cache
4425 labelled, 187 calls (10 failed), 0 from cache
4675 labelled, 197 calls (10 failed), 0 from cache
4750 labelled, 200 calls (10 failed), 0 from cache
this run: 200 calls, 10 failed, 4750 headlines labelled
6250 labelled headlines
year n bear irrel bull relevant mean xau
2009 10 0 10 0 0.0% +0.000
2010 9 0 9 0 0.0% +0.000
2011 10 0 10 0 0.0% +0.000
2012 2 0 2 0 0.0% +0.000
2015 13 0 13 0 0.0% +0.000
2016 6 0 6 0 0.0% +0.000
2017 30 0 30 0 0.0% +0.000
2018 19 0 19 0 0.0% +0.000
2019 462 23 359 58 17.5% +0.040
2020 507 29 366 69 19.3% +0.061
2021 638 45 496 65 17.2% +0.018
2022 707 72 477 97 23.9% +0.020
2023 796 55 574 108 20.5% +0.037
2024 807 48 568 139 23.2% +0.062
2025 1067 97 543 329 39.9% +0.118
2026 1167 218 559 258 40.8% +0.005
(relevant = |xau| >= 0.25, irrel = |xau| <= 0.10, the rest is the grey zone)