Spaces:
Running
Running
Commit History
-
Average 20 sampled paths; step daily forecasts by business days Running
Verified signaturesample_count=1 returned a single random draw from the predictive distribution rather than an estimate of its mean, which made the forecasts pure noise (next-day MAE ~2.5%% of price vs ~0.6%% for assuming no change, corr ~0.02). Now averages KRONOS_SAMPLE_COUNT=20 paths. Target dates were stepped with pd.date_range at the median bar delta (1 day), so 29%% of stored forecasts landed on Saturdays and Sundays. timeline.py now steps daily bars by business days when the input history has no weekend bars; crypto and intraday intervals keep calendar stepping. /health echoes the sampling config so the live build is identifiable.5299cb1
-
Initial Kronos Trading API deployment
Verified signature0ccee0a -
initial commit
Verified signature6f72567