app/engine/detection.py

import numpy as np
from transitleastsquares import transitleastsquares

def run_tls(time: np.ndarray, flux: np.ndarray, deep_recovery_mode: bool = False):
    """
    Run Transit Least Squares (TLS) to detect transits.
    Finds the strongest periodic transit signal.
    """
    model = transitleastsquares(time, flux)
    
    if deep_recovery_mode and len(time) > 100:
        baseline = time[-1] - time[0]
        # In deep recovery, force search up to exactly half the baseline
        # Standard TLS sometimes defaults lower based on heuristics
        results = model.power(period_max=baseline / 2.01, oversampling_factor=3, use_threads=4)
    else:
        # Fast survey mode: use defaults
        results = model.power(use_threads=4)
    
    # SDE (Signal Detection Efficiency) > 7.0 is typically considered a significant detection
    transit_detected = bool(results.SDE > 7.0)
    
    # Safely convert transit_times to list
    if results.transit_times is None:
        t_times = []
    elif isinstance(results.transit_times, list):
        t_times = results.transit_times
    else:
        t_times = results.transit_times.tolist()

    return {
        "transit_detected": transit_detected,
        "period": float(results.period),
        "depth": float(1.0 - results.depth),
        "duration": float(results.duration),
        "sde": float(results.SDE),
        "tls_confidence": float(min(100.0, results.SDE * 10.0)), # Scale SDE to a 0-100 score roughly
        "power_spectrum": {
            "periods": results.periods.tolist(),
            "power": results.power.tolist()
        },
        "transit_times": t_times
    }

from transitleastsquares import transit_mask

def run_multi_tls(time: np.ndarray, flux: np.ndarray, max_planets: int = 3):
    """
    Run iterative TLS to detect multiple planets.
    Masks out the transits of the strongest detected signal and searches again.
    Returns a list of candidate dictionaries.
    """
    candidates = []
    current_time = np.copy(time)
    current_flux = np.copy(flux)
    
    for i in range(max_planets):
        # Run TLS
        result = run_tls(current_time, current_flux)
        
        # Stop if no significant signal found
        if not result["transit_detected"]:
            break
            
        # Add candidate
        candidate = result.copy()
        candidate["candidate_number"] = i + 1
        candidates.append(candidate)
        
        # Mask out the detected transits for the next iteration
        if result["transit_times"]:
            t0 = result["transit_times"][0]
            # Create a boolean mask of in-transit points
            intransit = transit_mask(current_time, result["period"], result["duration"] * 1.5, t0)
            
            # Remove the in-transit points entirely to avoid TLS fitting to residuals
            valid_points = ~intransit
            current_time = current_time[valid_points]
            current_flux = current_flux[valid_points]
            
            if len(current_time) < 100:
                break # Too few points left
        else:
            break
            
    return candidates