Adaptive AI Polar Oscillator [by Oberlunar]

3 日前

Adaptive AI Oscillator blends trading signals with two order-flow style oscillators and a lightweight online-learning model to keep it reactive, adaptive and computationally feasible.




What it is

• A lightweight Multi Layer Perceptron (neural net) updates online on every bar, so it keeps adapting as conditions change.
  
• An adaptive collector that fuses features like Price (close, ohlc4, etc...), a selectable (but not used in the original implementation) Moving Average (EMA/SMA/WMA/RMA/HMA/DEMA/TEMA), RSI, the classic volume datafeeds, plus two “OberPolar” oscillators computed above and below the current integral area price.
  
  

What you see

• White line — the model’s denormalised forecast (in price units).
  
• Colored price line — actual price, shown aqua when forecast ≥ price (“golden” bias) and red when forecast < price (“death” bias).
  

Why it helps

• Combines heterogeneous information (trend, momentum, participation, regional buy/sell pressure) into a single adaptive forecast.
  
• Online learning reduces regime staleness versus fixed-parameter indicators.
  
• The aqua/red bias offers a quick, visual state for discretionary decisions.
  

How it works (intuitive)

1. Each AI input is standardised (z-score) with optional clamping to mitigate outliers.
   
2. A rolling window of recent values feeds a 2-layer AI to predict one step ahead.
   
3. After each bar closes, the model compares forecast vs. reality and nudges its weights (SGD with momentum, L2, optional gradient clipping).
   
4. The forecast is de-standardised back to price units and plotted as the white line.
   

Reading guide

• Crossovers between forecast and price often mark potential bias flips.
  
• Persistent aqua → model perceives supportive/positive conditions.
  
• Persistent red → model perceives headwinds/negative conditions.
  

Complex Strategy — Oscillator Trendline Break
Connect the first pivot in the fading bias with the first pivot in the new bias, then trade the break of that line in the direction of the new bias.

Idea in one line

• Use the Adaptive AI Oscillator (green = bullish bias, red = bearish). When bias flips, build a line across the oscillator pivots that “span” the transition; the break of that line times the entry.
  

Long setup (mirror for shorts)

1. Bias transition: a bearish (red) regime is ongoing, then the oscillator turns bullish (green).
   
2. Anchor pivots: take the first MIN in red just before/around the flip and the first MAX in green after the flip. Draw a trendline L through these two oscillator values (time–value line).
   
3. Trigger: enter LONG on the close that breaks above L—optional confirmations: price above your MA, non-decreasing volume, no immediate supply zone overhead.
   
4. Risk: stop below the last oscillator swing low or below a retest of L; first target at 1R–1.5R or at the opposite bias zone; trail under successive oscillator higher lows.
   

Short setup

1. Bias turns from green (bullish) to red (bearish).
   
2. Connect the first MAX in green to the first MIN in red → line L.
   
3. Enter SHORT on a close below L; stop above the last oscillator swing high; symmetric targets/trailing.
   

Complex Strategy #2 — Bias-Pivot Breakout with Exit on Line Failure
Connect two pivots of the same bias to build a dynamic barrier; trade the breakout in the bias direction and exit when that line later fails.

Long play (mirror for shorts)

1. Build the line. During a green (bullish) phase, mark the first two local MAX of the oscillator. Connect them to form the yellow resistance line L (extend it right). If a new, clearer MAX appears before a break, re-anchor using the two most recent highs.
   
2. Entry trigger. Go LONG on a close above L (the “Break and LONG” in the image). Optional filters: price above your MA, rising volume, no immediate overhead level.
   
3. Risk. Initial stop: below the last oscillator swing low or below the retest of L. Position size for 1–2R baseline.
   
4. Exit. Close the long when the oscillator later breaks back below L (the “Break and LONG exit”), or on a bias flip to red, or at a fixed target/trailing under higher lows.
   

Short play (symmetric)

1. In a red phase, connect the first two local MIN to form support line L.
   
2. Enter SHORT on a close below L; stop above the last oscillator swing high; exit on a break back above L or on a flip to green.
   

Notes

• Require a minimum slope/spacing between pivots to avoid flat/noisy lines.
  
• Re-anchor the line if fresher pivots emerge before a valid break.
  
• Use with your regime filter (MA slope, higher-timeframe bias) to reduce whipsaws.
  

Complex Strategy #3 — Lateral Box & Zero-Slope Breakout
An easy way to understand sideways phases and the next price direction: draw two zero-slope lines (flat upper/lower bounds) across the oscillator’s lateral area; when a strong break occurs, trade in the direction of that break.

How to use it

1. Identify a lateral area on the oscillator (flat, low-variance region). Place a flat upper line on tops and a flat lower line on bottoms (slope ≈ 0).
   
2. Wait for a decisive break: close outside the band with expansion (range/true range rising, or a wide candle).
   • Break up → bias for LONG.
   • Break down → bias for SHORT.
   

Why it helps

• Flat lines isolate congestion; the next impulsive move is often revealed by which side is broken with force.
  
• It filters noise inside the range and focuses attention on the transition from balance → imbalance.
  

Practical filters (optional)

• Require minimum bar body/ATR on the breakout candle to avoid false breaks.
  
• Confirm with your regime filter (e.g., price above/below your MA) or a quick retest that holds.
  
• Invalidate the signal if the price immediately returns inside the band on the next bar.
  

General Operational notes

• If new pivots form before a break, re-anchor the line with the most recent qualifying pair (keeps the structure fresh).
  
• Ignore very shallow lines (near-flat): require a minimum slope or angle to avoid noise.
  
• Combine with your bias filter (e.g., MA slope/regime) to reduce false starts.
  

Limits & good practice

• Adaptive models can react to noise; treat signals as context within a risk-managed plan.
  
• No model predicts the future—this summarises evolving conditions compactly.
  

— Oberlunar 👁 ★

3 日前

リリースノート

small fix

3 日前

リリースノート

small fix