AlphaEar Dashboard — Ensemble Forecasting, Alerts, and Big UI Refinements

projectupdatefinanceaipython
...

Since the initial launch in May, AlphaEar Dashboard has picked up a fair bit of polish.

Ensemble forecasting

The earlier linear-regression approach was prone to spurious trends. The new approach is a 3-model ensemble blending naive persistence, log-return drift, and EMA-trend extrapolation, smooth-weighted with a simple average. An EMA-dampened stability smoother (α=0.4) prevents flip-flopping between consecutive forecast runs. The forecast store switched from INSERT OR REPLACE to INSERT so historical runs are immutable and honest MAPE calculation is possible.

Walk-forward backtesting

A new scripts/backtest_walkforward.py harness simulates realistic out-of-sample accuracy. Current figures for AAPL: MAPE 2.45%, MASE 1.057. Volatility regime detection flags periods where ATR/price exceeds 3% as HIGH VOL.

Alerts dashboard

A dedicated Alerts tab (keyboard shortcut 7) shows real-time market alerts with mobile swipe gestures, alongside a portfolio chart for a quick position overview.

UI refinements

  • About tab (8): comprehensive project overview, replaces the verbose mini-section in Settings
  • Login page: submit spinner, password visibility toggle, focus ring, autofill styling, logo pulse animation, skeleton loading screen
  • Lazy loading: tabs load on demand, no unnecessary JS churn on initial page load
  • Accuracy view: MAPE summary cards with inline sparklines

All live at alphaear.oilygold.xyz. Open source on GitHub.