gahusb/ai-trade
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(signal_v2-phase3b): force FP32 + predict_quantiles positional args

Browse Source 
ChronosBoltPipeline.predict_quantiles takes `inputs` positional, not
`context` keyword. Use positional with TypeError fallback for older
chronos versions.

FP16 caused inf overflow on Korean stock prices (e.g. 280,000원 >
FP16 max 65,504). Force FP32 for prices to avoid this. Chronos model
itself handles internal scaling.

Verified end-to-end: 60-day daily fetch → Chronos predict → quantile
output. Example 005930: median=-0.59%, q10=-8.9%, q90=+6.4%, conf=0.0
(low conf is mathematically correct when median is near zero relative
to distribution width).

45/45 tests still pass.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
gahusb
2026-05-17 09:12:10 +09:00
parent 8eefe9d79d
commit 760f914d3b
1 changed files with 16 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
signal_v2/chronos_predictor.py
View File

@@ -35,7 +35,9 @@ class ChronosPredictor:
pipeline_cls = ChronosPipeline pipeline_cls = ChronosPipeline
self._device = device or ("cuda" if torch.cuda.is_available() else "cpu") self._device = device or ("cuda" if torch.cuda.is_available() else "cpu")
dtype = torch.float16 if self._device == "cuda" else torch.float32 # Always use float32 — Korean stock prices (e.g. 280,000원) exceed FP16 max (~65,504)
# causing inf in quantile output. FP32 is safe for typical price magnitudes.
dtype = torch.float32
logger.info("Loading Chronos pipeline: %s on %s (cls=%s)", logger.info("Loading Chronos pipeline: %s on %s (cls=%s)",
model_name, self._device, pipeline_cls.__name__) model_name, self._device, pipeline_cls.__name__)
# Try `dtype` (newer API) first, fall back to `torch_dtype` (older) # Try `dtype` (newer API) first, fall back to `torch_dtype` (older)
@@ -76,6 +78,14 @@ class ChronosPredictor:
# Modern API: predict_quantiles (ChronosBolt / Chronos-2) # Modern API: predict_quantiles (ChronosBolt / Chronos-2)
if hasattr(self._pipeline, "predict_quantiles"): if hasattr(self._pipeline, "predict_quantiles"):
quantile_levels = [0.1, 0.5, 0.9] quantile_levels = [0.1, 0.5, 0.9]
# ChronosBolt API: positional `inputs` (first arg). Older variants use `context`.
try:
quantiles_tensor, _ = self._pipeline.predict_quantiles(
contexts,
prediction_length=prediction_length,
quantile_levels=quantile_levels,
)
except TypeError:
quantiles_tensor, _ = self._pipeline.predict_quantiles( quantiles_tensor, _ = self._pipeline.predict_quantiles(
context=contexts, context=contexts,
prediction_length=prediction_length, prediction_length=prediction_length,