Files
QuanTAlib/lib/trends_IIR/qema/Qema.cs
T
86fe32a682 SIMD Refactor: Merge simd-dev into dev (#55)
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
Co-authored-by: aider (openrouter/anthropic/claude-sonnet-4) <aider@aider.chat>
Co-authored-by: Warp <agent@warp.dev>
2026-01-18 19:02:03 -08:00

559 lines
18 KiB
C#

using System.Runtime.CompilerServices;
using System.Runtime.InteropServices;
namespace QuanTAlib;
/// <summary>
/// QEMA: Quad Exponential Moving Average with Progressive Alphas and Zero-Lag Weighting
/// </summary>
/// <remarks>
/// QEMA uses four cascaded EMAs with progressively increasing alphas (decreasing responsiveness)
/// and combines them using optimized weights that minimize energy while achieving zero DC lag.
///
/// Calculation:
/// 1. Base alpha: α₁ = 2 / (period + 1)
/// 2. Progressive alphas: r = (1/α₁)^(1/4), then α₂ = α₁·r, α₃ = α₂·r, α₄ = α₃·r
/// 3. Four cascaded EMAs: EMA1(input), EMA2(EMA1), EMA3(EMA2), EMA4(EMA3)
/// 4. Cumulative lags: L₁ = (1-α₁)/α₁, L₂ = L₁ + (1-α₂)/α₂, etc.
/// 5. Option A weights: Minimize energy subject to Σw=1 and Σw·L=0 (zero DC lag)
/// 6. Output: w₁·EMA1 + w₂·EMA2 + w₃·EMA3 + w₄·EMA4
///
/// O(1) update:
/// Uses four EMA state accumulators, each with O(1) update complexity.
///
/// IsHot:
/// Becomes true when the slowest EMA (stage 1) has converged to within 5% coverage.
/// </remarks>
[SkipLocalsInit]
public sealed class Qema : AbstractBase
{
[StructLayout(LayoutKind.Auto)]
private record struct EmaState(double Ema, double E, bool IsHot, bool IsCompensated)
{
public static EmaState New() => new() { Ema = 0, E = 1.0, IsHot = false, IsCompensated = false };
}
private readonly double _alpha1, _alpha2, _alpha3, _alpha4;
private readonly double _decay1, _decay2, _decay3, _decay4;
private EmaState _state1 = EmaState.New();
private EmaState _state2 = EmaState.New();
private EmaState _state3 = EmaState.New();
private EmaState _state4 = EmaState.New();
private EmaState _p_state1 = EmaState.New();
private EmaState _p_state2 = EmaState.New();
private EmaState _p_state3 = EmaState.New();
private EmaState _p_state4 = EmaState.New();
private readonly TValuePublishedHandler _handler;
private double _lastValidValue;
private double _p_lastValidValue;
private const double COVERAGE_THRESHOLD = 0.05;
private const double COMPENSATOR_THRESHOLD = 1e-10;
private const double DEGENERATE_THRESHOLD = 1e-12;
/// <summary>
/// True when the slowest EMA (stage 1) has warmed up and is providing valid results.
/// </summary>
public override bool IsHot => _state1.E <= COVERAGE_THRESHOLD;
/// <summary>
/// Creates QEMA with specified period.
/// Alpha1 = 2 / (period + 1), with progressive alphas ramped geometrically.
/// </summary>
/// <param name="period">Period for base EMA calculation (must be > 0)</param>
public Qema(int period)
{
ArgumentOutOfRangeException.ThrowIfNegativeOrZero(period);
_alpha1 = Clamp01(2.0 / (period + 1));
// Progressive alpha ramp: r = (1/α₁)^(1/4) → α₂=α₁^(3/4), α₃=α₁^(1/2), α₄=α₁^(1/4)
double r = Math.Pow(1.0 / _alpha1, 0.25);
_alpha2 = Clamp01(_alpha1 * r);
_alpha3 = Clamp01(_alpha2 * r);
_alpha4 = Clamp01(_alpha3 * r);
_decay1 = 1.0 - _alpha1;
_decay2 = 1.0 - _alpha2;
_decay3 = 1.0 - _alpha3;
_decay4 = 1.0 - _alpha4;
Name = $"Qema({period})";
WarmupPeriod = period;
_handler = Handle;
}
/// <summary>
/// Creates QEMA with specified source and period.
/// Subscribes to source.Pub event.
/// </summary>
/// <param name="source">Source to subscribe to</param>
/// <param name="period">Period for base EMA calculation</param>
public Qema(ITValuePublisher source, int period) : this(period)
{
source.Pub += _handler;
}
/// <summary>
/// Creates QEMA with specified source TSeries and period.
/// Primes with historical data and subscribes to updates.
/// </summary>
/// <param name="source">Source TSeries</param>
/// <param name="period">Period for base EMA calculation</param>
public Qema(TSeries source, int period) : this(period)
{
Prime(source.Values);
if (source.Count > 0)
{
Last = new TValue(source.LastTime, Last.Value);
}
source.Pub += _handler;
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
private static double Clamp01(double x) => Math.Min(1.0, Math.Max(x, DEGENERATE_THRESHOLD));
[MethodImpl(MethodImplOptions.AggressiveInlining)]
private static double Lag(double alpha) => (1.0 - alpha) / alpha;
private void Handle(object? sender, in TValueEventArgs e) => Update(e.Value, e.IsNew);
/// <summary>
/// Initializes the indicator state using the provided history.
/// </summary>
/// <param name="source">Historical data</param>
/// <param name="step">Optional time step (not used)</param>
public override void Prime(ReadOnlySpan<double> source, TimeSpan? step = null)
{
if (source.Length == 0) return;
// Reset state
_state1 = EmaState.New();
_state2 = EmaState.New();
_state3 = EmaState.New();
_state4 = EmaState.New();
_p_state1 = EmaState.New();
_p_state2 = EmaState.New();
_p_state3 = EmaState.New();
_p_state4 = EmaState.New();
_lastValidValue = 0;
_p_lastValidValue = 0;
int len = source.Length;
double lastValid = 0;
// Find first finite value
for (int i = 0; i < len; i++)
{
if (double.IsFinite(source[i]))
{
lastValid = source[i];
break;
}
}
EmaState s1 = _state1;
EmaState s2 = _state2;
EmaState s3 = _state3;
EmaState s4 = _state4;
for (int i = 0; i < len; i++)
{
double val = source[i];
if (double.IsFinite(val))
lastValid = val;
else
val = lastValid;
double e1 = ComputeEma(val, _alpha1, _decay1, ref s1);
double e2 = ComputeEma(e1, _alpha2, _decay2, ref s2);
double e3 = ComputeEma(e2, _alpha3, _decay3, ref s3);
ComputeEma(e3, _alpha4, _decay4, ref s4);
}
_state1 = s1;
_state2 = s2;
_state3 = s3;
_state4 = s4;
_lastValidValue = lastValid;
// Calculate final output
double e1_final = GetCompensated(_state1);
double e2_final = GetCompensated(_state2);
double e3_final = GetCompensated(_state3);
double e4_final = GetCompensated(_state4);
var (w1, w2, w3, w4) = ComputeWeights(_alpha1, _alpha2, _alpha3, _alpha4);
double result = Math.FusedMultiplyAdd(w1, e1_final, Math.FusedMultiplyAdd(w2, e2_final, Math.FusedMultiplyAdd(w3, e3_final, w4 * e4_final)));
Last = new TValue(DateTime.MinValue, result);
_p_state1 = _state1;
_p_state2 = _state2;
_p_state3 = _state3;
_p_state4 = _state4;
_p_lastValidValue = _lastValidValue;
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
private static double GetCompensated(EmaState s)
{
if (s.IsCompensated) return s.Ema;
return s.Ema / (1.0 - s.E);
}
[MethodImpl(MethodImplOptions.AggressiveInlining | MethodImplOptions.AggressiveOptimization)]
public override TValue Update(TValue input, bool isNew = true)
{
if (isNew)
{
_p_state1 = _state1;
_p_state2 = _state2;
_p_state3 = _state3;
_p_state4 = _state4;
_p_lastValidValue = _lastValidValue;
}
else
{
_state1 = _p_state1;
_state2 = _p_state2;
_state3 = _p_state3;
_state4 = _p_state4;
_lastValidValue = _p_lastValidValue;
}
double val = input.Value;
if (double.IsFinite(val))
_lastValidValue = val;
else
val = _lastValidValue;
// Cascaded EMAs
double e1 = ComputeEma(val, _alpha1, _decay1, ref _state1);
double e2 = ComputeEma(e1, _alpha2, _decay2, ref _state2);
double e3 = ComputeEma(e2, _alpha3, _decay3, ref _state3);
double e4 = ComputeEma(e3, _alpha4, _decay4, ref _state4);
// Compute weights and combine
var (w1, w2, w3, w4) = ComputeWeights(_alpha1, _alpha2, _alpha3, _alpha4);
double result = Math.FusedMultiplyAdd(w1, e1, Math.FusedMultiplyAdd(w2, e2, Math.FusedMultiplyAdd(w3, e3, w4 * e4)));
Last = new TValue(input.Time, result);
PubEvent(Last, isNew);
return Last;
}
public override TSeries Update(TSeries source)
{
if (source.Count == 0) return [];
int len = source.Count;
List<long> t = new(len);
List<double> v = new(len);
CollectionsMarshal.SetCount(t, len);
CollectionsMarshal.SetCount(v, len);
var tSpan = CollectionsMarshal.AsSpan(t);
var vSpan = CollectionsMarshal.AsSpan(v);
source.Times.CopyTo(tSpan);
var sourceValues = source.Values;
EmaState s1 = _state1;
EmaState s2 = _state2;
EmaState s3 = _state3;
EmaState s4 = _state4;
double lastValid = _lastValidValue;
var (w1, w2, w3, w4) = ComputeWeights(_alpha1, _alpha2, _alpha3, _alpha4);
for (int i = 0; i < len; i++)
{
double val = sourceValues[i];
if (double.IsFinite(val))
lastValid = val;
else
val = lastValid;
double e1 = ComputeEma(val, _alpha1, _decay1, ref s1);
double e2 = ComputeEma(e1, _alpha2, _decay2, ref s2);
double e3 = ComputeEma(e2, _alpha3, _decay3, ref s3);
double e4 = ComputeEma(e3, _alpha4, _decay4, ref s4);
vSpan[i] = Math.FusedMultiplyAdd(w1, e1, Math.FusedMultiplyAdd(w2, e2, Math.FusedMultiplyAdd(w3, e3, w4 * e4)));
}
_state1 = s1;
_state2 = s2;
_state3 = s3;
_state4 = s4;
_p_state1 = s1;
_p_state2 = s2;
_p_state3 = s3;
_p_state4 = s4;
_lastValidValue = lastValid;
_p_lastValidValue = lastValid;
Last = new TValue(tSpan[len - 1], vSpan[len - 1]);
return new TSeries(t, v);
}
/// <summary>
/// Computes the bias-corrected EMA value and updates state.
/// </summary>
[MethodImpl(MethodImplOptions.AggressiveInlining)]
private static double ComputeEma(double input, double alpha, double decay, ref EmaState state)
{
state.Ema = Math.FusedMultiplyAdd(state.Ema, decay, alpha * input);
double result;
if (!state.IsCompensated)
{
state.E *= decay;
if (!state.IsHot && state.E <= COVERAGE_THRESHOLD)
state.IsHot = true;
if (state.E <= COMPENSATOR_THRESHOLD)
{
state.IsCompensated = true;
result = state.Ema;
}
else
{
result = state.Ema / (1.0 - state.E);
}
}
else
{
result = state.Ema;
}
return result;
}
/// <summary>
/// Computes Option A weights for minimum energy with zero DC lag constraint.
/// Solves: Σw = 1, Σw·L = 0 (δ=0 for zero lag)
/// </summary>
[MethodImpl(MethodImplOptions.AggressiveInlining)]
private static (double w1, double w2, double w3, double w4) ComputeWeights(double a1, double a2, double a3, double a4)
{
double t1 = Lag(a1);
double t2 = Lag(a2);
double t3 = Lag(a3);
double t4 = Lag(a4);
// Cumulative lags
double L1 = t1;
double L2 = t1 + t2;
double L3 = L2 + t3;
double L4 = L3 + t4;
// Option A: min-energy with constraints Σw=1, Σw·L=δ (δ=0)
double B = L1 + L2 + L3 + L4;
double C = Math.FusedMultiplyAdd(L1, L1, Math.FusedMultiplyAdd(L2, L2, Math.FusedMultiplyAdd(L3, L3, L4 * L4)));
double D = Math.FusedMultiplyAdd(4.0, C, -B * B);
double w1, w2, w3, w4;
if (Math.Abs(D) < DEGENERATE_THRESHOLD)
{
// Degenerate case (e.g., alpha=1 → all L=0): output is EMA1≈input
w1 = 1.0;
w2 = 0.0;
w3 = 0.0;
w4 = 0.0;
}
else
{
double lambda = C / D;
double mu = -B / D;
w1 = Math.FusedMultiplyAdd(mu, L1, lambda);
w2 = Math.FusedMultiplyAdd(mu, L2, lambda);
w3 = Math.FusedMultiplyAdd(mu, L3, lambda);
w4 = Math.FusedMultiplyAdd(mu, L4, lambda);
}
return (w1, w2, w3, w4);
}
/// <summary>
/// Calculates QEMA for the entire series using a new instance.
/// </summary>
/// <param name="source">Input series</param>
/// <param name="period">QEMA period</param>
/// <returns>QEMA series</returns>
public static TSeries Batch(TSeries source, int period)
{
var qema = new Qema(period);
return qema.Update(source);
}
/// <summary>
/// Calculates QEMA in-place using period, writing results to pre-allocated output span.
/// Zero-allocation method for maximum performance.
/// </summary>
/// <param name="source">Input values</param>
/// <param name="output">Output span (must be same length as source)</param>
/// <param name="period">QEMA period (must be > 0)</param>
[MethodImpl(MethodImplOptions.AggressiveOptimization)]
public static void Batch(ReadOnlySpan<double> source, Span<double> output, int period)
{
if (source.Length != output.Length)
throw new ArgumentException("Source and output must have the same length", nameof(output));
ArgumentOutOfRangeException.ThrowIfNegativeOrZero(period);
if (source.Length == 0) return;
double alpha1 = Clamp01(2.0 / (period + 1));
double r = Math.Pow(1.0 / alpha1, 0.25);
double alpha2 = Clamp01(alpha1 * r);
double alpha3 = Clamp01(alpha2 * r);
double alpha4 = Clamp01(alpha3 * r);
double decay1 = 1.0 - alpha1;
double decay2 = 1.0 - alpha2;
double decay3 = 1.0 - alpha3;
double decay4 = 1.0 - alpha4;
double lastValid = 0;
// Find first finite value
for (int i = 0; i < source.Length; i++)
{
if (double.IsFinite(source[i]))
{
lastValid = source[i];
break;
}
}
// EMA states
double ema1_val = 0, ema1_e = 1.0;
bool ema1_compensated = false;
double ema2_val = 0, ema2_e = 1.0;
bool ema2_compensated = false;
double ema3_val = 0, ema3_e = 1.0;
bool ema3_compensated = false;
double ema4_val = 0, ema4_e = 1.0;
bool ema4_compensated = false;
var (w1, w2, w3, w4) = ComputeWeights(alpha1, alpha2, alpha3, alpha4);
for (int i = 0; i < source.Length; i++)
{
double val = source[i];
if (double.IsFinite(val))
lastValid = val;
else
val = lastValid;
// EMA1
ema1_val = Math.FusedMultiplyAdd(ema1_val, decay1, alpha1 * val);
double e1;
if (!ema1_compensated)
{
ema1_e *= decay1;
if (ema1_e <= COMPENSATOR_THRESHOLD)
{
ema1_compensated = true;
e1 = ema1_val;
}
else
{
e1 = ema1_val / (1.0 - ema1_e);
}
}
else
{
e1 = ema1_val;
}
// EMA2
ema2_val = Math.FusedMultiplyAdd(ema2_val, decay2, alpha2 * e1);
double e2;
if (!ema2_compensated)
{
ema2_e *= decay2;
if (ema2_e <= COMPENSATOR_THRESHOLD)
{
ema2_compensated = true;
e2 = ema2_val;
}
else
{
e2 = ema2_val / (1.0 - ema2_e);
}
}
else
{
e2 = ema2_val;
}
// EMA3
ema3_val = Math.FusedMultiplyAdd(ema3_val, decay3, alpha3 * e2);
double e3;
if (!ema3_compensated)
{
ema3_e *= decay3;
if (ema3_e <= COMPENSATOR_THRESHOLD)
{
ema3_compensated = true;
e3 = ema3_val;
}
else
{
e3 = ema3_val / (1.0 - ema3_e);
}
}
else
{
e3 = ema3_val;
}
// EMA4
ema4_val = Math.FusedMultiplyAdd(ema4_val, decay4, alpha4 * e3);
double e4;
if (!ema4_compensated)
{
ema4_e *= decay4;
if (ema4_e <= COMPENSATOR_THRESHOLD)
{
ema4_compensated = true;
e4 = ema4_val;
}
else
{
e4 = ema4_val / (1.0 - ema4_e);
}
}
else
{
e4 = ema4_val;
}
output[i] = Math.FusedMultiplyAdd(w1, e1, Math.FusedMultiplyAdd(w2, e2, Math.FusedMultiplyAdd(w3, e3, w4 * e4)));
}
}
/// <summary>
/// Resets the QEMA state.
/// </summary>
public override void Reset()
{
_state1 = EmaState.New();
_state2 = EmaState.New();
_state3 = EmaState.New();
_state4 = EmaState.New();
_p_state1 = EmaState.New();
_p_state2 = EmaState.New();
_p_state3 = EmaState.New();
_p_state4 = EmaState.New();
_lastValidValue = 0;
_p_lastValidValue = 0;
Last = default;
}
}