Overall Statistics
from Algorithms.StandardAlgorithm import StandardAlgorithm
from Alphas.RsiAlphaModel import RsiAlphaModel
from Execution.StandardDeviationExecutionModel import StandardDeviationExecutionModel

class MainAlgorithm(StandardAlgorithm):
    Timeframes = [1, 5, 15, 30, 60, 240, 390, 1950]

    def Initialize(self):
        self.SetExecution(StandardDeviationExecutionModel(60, 2, Resolution.Minute))
        self.SetAlphas(RsiAlphaModel)
        
        # Securities traded
        symbols = [ Symbol.Create("CRON", SecurityType.Equity, Market.USA) ]
        self.SetUniverseSelection( ManualUniverseSelectionModel(symbols) )
import math
import numpy as np
from Indicators.StandardIndicator import StandardIndicator

# def EhlerRSI(symbol, **kwargs):
#     _symbol = symbol if isinstance(symbol, str) else symbol.Value
#     RegisterIndicator(_symbol, EhlerRsiDiscriminator(**kwargs), kwargs['consolidator'].Period);
#     return indicator


class EhlerRsiDiscriminator(StandardIndicator):
    Warmup = 7
    Name = "Ehler's RSI + Discriminator"
    
    
    def __init__(self, double_smoothing=True, roofing_upper=48, roofing_lower=10, overbought=70, oversold=30, **kwargs):
        super().__init__()
        self.DoubleSmoothing = double_smoothing
        self.RoofingUpper = roofing_upper
        self.RoofingLower = roofing_lower
        self.Overbought = overbought
        self.Oversold = oversold
        self.Source = RollingWindow[Decimal](3)
        self.Filter = RollingWindow[float](3)
        self.Highpass = RollingWindow[float](3)
        self.Period = RollingWindow[Decimal](2)
        self.SmoothedPeriod = RollingWindow[float](2)
        self.Smooth = RollingWindow[float](7)
        self.Detrend = RollingWindow[float](7)
        self.HL2 = RollingWindow[Decimal](4)
        self.I1 = RollingWindow[float](7)
        self.I2 = RollingWindow[float](7)
        self.Q1 = RollingWindow[float](7)
        self.Q2 = RollingWindow[float](7)
        self.Re = RollingWindow[float](2)
        self.Im = RollingWindow[float](2)
        self.MA = RollingWindow[float](2)
        self.RoofedSource = RollingWindow[float](2)
        
        # Start with history
        self.FillWindows(self.Filter, self.Highpass, self.Period, self.Smooth, self.Detrend, self.I1, self.I2, self.Q1, self.Q2, self.Re, self.Im, self.MA, self.RoofedSource)
        self.FillWindows(self.SmoothedPeriod, value=6)

    def __repr__(self):
        out = ""
        for attribute, data in vars(self).items():
            out += "{}: {}, ".format(attribute, (data[0] if isinstance(data, self.Source.__class__) or isinstance(data, self.Filter.__class__) else data))
        return out
    
    
    # Rolling Moving Average (or Wells Wilders MA)
    def WWma(self, src, l):
        self.MA.Add(np.nan_to_num((self.MA[1] * (l - 1) + src) / l))
        return self.MA[0]
    
    
    # RSI function.
    def EhRSI(self, l):
        change = self.RoofedSource[0] - self.RoofedSource[1]
        up = self.WWma(max(change, 0), l)
        down = self.WWma(-min(change, 0), l)
        return 100 if down == 0 else 0 if up == 0 else 100 - (100 / (1 + up / down))
    
    
    def EhlersSuperSmootherFilter(self):
        f = (math.pi * math.sqrt(2)) / self.RoofingLower
        a = math.exp(-f)
        c3 = -a * a
        c2 = 2 * a * math.cos(f)
        c1 = 1 - c2 - c3
        self.Filter.Add(c1 * (self.Highpass[0] + self.Highpass[1]) / 2 + c2 * self.Filter[1] + c3 * self.Filter[2])
        return self.Filter[0]
        

    def EhlersRoofingFilter(self):
        f = math.sqrt(2) * math.pi / self.RoofingUpper
        alpha1 = (math.cos(f) + math.sin(f) - 1) / math.cos(f)
        self.Highpass.Add(np.nan_to_num(pow(1 - alpha1 / 2, 2) * (self.Source[0] - 2 * self.Source[1] + self.Source[2]) + 2 * (1 - alpha1) * self.Highpass[1] - pow(1 - alpha1, 2) * self.Highpass[2]))
        return self.EhlersSuperSmootherFilter() if self.DoubleSmoothing else self.Highpass[0]
    
    
    def Quadrature(self, src):
        C1 = 0.0962
        C2 = 0.5769
        C3 = (self.Period[1] * 0.075 + 0.54)
        return (src[0] * C1 + src[2] * C2 - src[4] * C2 - src[6] * C1) * C3
    
    
    def Update(self, data):
        self.Source.Add(data.Close)
        self.HL2.Add((data.High + data.Low) / 2)
        self.IsReady = self.Source.IsReady and self.HL2.IsReady
        if not self.IsReady: return
        
        #
        # --- Start the Homodyne Discriminator Caculations
        #
        # Mutable Variables (non-series)
        self.Smooth.Add((self.HL2[0] * 4.0 + self.HL2[1] * 3.0 + self.HL2[2] * 2.0 + self.HL2[3]) / 10.0)
        self.Detrend.Add(self.Quadrature(self.Smooth))
        
        # Compute InPhase and Quadrature components
        self.Q1.Add(self.Quadrature(self.Detrend))
        self.I1.Add(self.Detrend[3])
        
        # Advance Phase of I1 and Q1 by 90 degrees
        jI = self.Quadrature(self.I1)
        jQ = self.Quadrature(self.Q1)
        
        # Phaser addition for 3 bar averaging and 
        # Smooth i and q components before applying discriminator
        self.I2.Add(0.2 * (self.I1[0] - jQ) + 0.8 * self.I2[1])
        self.Q2.Add(0.2 * (self.Q1[0] + jI) + 0.8 * self.Q2[1])
        
        # Extract Homodyne Discriminator
        self.Re.Add(0.2 * (self.I2[0] * self.I2[1] + self.Q2[0] * self.Q2[1]) + 0.8 * self.Re[1])
        self.Im.Add(0.2 * (self.I2[0] * self.Q2[1] - self.Q2[0] * self.I2[1]) + 0.8 * self.Im[1])
        
        Dp = ((2 * math.pi) / math.atan(self.Im[0] / self.Re[0])) if (self.Re[0] != 0 and self.Im[0] != 0) else 0
        Dp = max(max(min(min(Dp, 1.5 * self.Period[1]), 50), 0.6667 * self.Period[1]), 6)
        self.Period.Add(Dp * 0.2 + self.Period[1] * 0.8)
        self.SmoothedPeriod.Add(0.33 * self.Period[0] + self.SmoothedPeriod[1] * 0.67)
        
        self.RoofedSource.Add(self.EhlersRoofingFilter())
        self.Value = round(self.EhRSI(round(self.SmoothedPeriod[0] * 0.5 - 1)), 4)
        # self.Value = self.RoofedSource[0]
        # self.Value = IndicatorExtensions.Of(RelativeStrengthIndex(self.SmoothedPeriod[0] * 0.5 - 1, MovingAverageType.Wilders), self).Value
        return self.Value
from Utilities.Unsorted import *

class StandardAlphaModel(AlphaModel):
    Name = "Null Alpha"
    Indicator = None
    
    def __init__(self, algorithm):
        self.symbolData = {}
        self.Algorithm = algorithm
        self.Chart = Chart(self.Name, ChartType.Stacked)
        for consolidator in self.Algorithm.Consolidators:
            consolidator.DataConsolidated += self.Update
    
    
    # # S curve starting at 1 heading towards 0.
    # # This means when the oscillator is at 0, trend
    # # confidence is at its highest (most likely to reverse)
    # def TrendConfidence(self, x):
    #     return -0.5 * (1 + math.sin((math.pi * x) - (math.pi / 2))) + 1

    
    # # Equidistant thresholds within 0-1 range, ex; 3 means
    # # (1 at 16.7%, 50% and 83.3%)
    # # (0 at 0%, 33%, 66%, 100%)
    # def ThresholdConfidence(self, x, t=3, low=0):
    #     if low > 1: lowest_confidence = 1
    #     if low < 0: lowest_confidence = 0
    #     return abs(math.sin(t * x * math.pi)) + lowest_confidence
    
    
    # # Limit the threshold confidence to the trend confidence
    # def TrendLimitedThresholdConfidence(self, x, t=3):
    #     trend_confidence = this.TrendConfidence(x)
    #     threshold_confidence = this.ThresholdConfidence(x, t)
    #     return threshold_confidence if threshold_confidence <= trend_confidence else trend_confidence
    
    
    # This Update method somehow doesn't accept algorithm; only data
    def Update(self, algorithm, data):
        insights = []
        
        for symbol, sd in self.symbolData.items():
            for consolidator, indicator in sd.Indicators.items():
                if indicator.IsReady:
                    self.Algorithm.Plot(self.Name, ConsolidatedSymbolLabel(symbol, consolidator), indicator.Value)
                    # sd.Algorithm.Debug("{} {}: {}".format(data.Time, ConsolidatedSymbolLabel(symbol, consolidator), indicator.Source[0]))
        
        return insights
    
    
    def OnSecuritiesChanged(self, algorithm, changes):
        # None of my consolidators are present!
        for consolidator in algorithm.Consolidators:
            algorithm.Debug(consolidator.Period)
        # Nor are they found here!
        for consolidator in self.Algorithm.Consolidators:
            algorithm.Debug(consolidator.Period)
        for added in changes.AddedSecurities:
            self.symbolData[added.Symbol] = SymbolData(self, added)
            
        for removed in changes.RemovedSecurities:
            del self.symbolData[removed.Symbol]


class SymbolData:
    def __init__(self, alpha, security):
        self.Indicators = {}
        # So as a result, none of my indicators are created!
        for consolidator in alpha.Algorithm.Consolidators:
            indicator = alpha.Indicator()
            alpha.Algorithm.RegisterIndicator(security.Symbol, indicator, consolidator)
            alpha.Chart.AddSeries(Series(ConsolidatedSymbolLabel(security.Symbol, consolidator), SeriesType.Line))
            self.Indicators[consolidator] = indicator
from Alphas.StandardAlphaModel import *
from Indicators.EhlerRsiDiscriminator import EhlerRsiDiscriminator

class RsiAlphaModel(StandardAlphaModel):
    Indicator = EhlerRsiDiscriminator
    Name = EhlerRsiDiscriminator.Name
    
    
    # See this Update method accepts algorithm and data
    def Update(self, algorithm, data):
        insights = super().Update(algorithm, data)
        
        # data.Period - contains the timestamp representing the consolidated timeframe the data applies to.
        # We need to introduce MTF logic in aggregating confidence. Should we be keeping RollingWindow data
        # in the AlphaModel
        # for underlying in data:
        #     c = self.TrendLimitedThresholdConfidence(data[underlying], 3)
        #     time_delta = minutes * self.TrendConfidence()
        #     insights.append( Insight(data[underlying], time_delta, InsightType.Price, InsightDirection.Up, confidence=c) )
        #     insights.append( Insight(data[underlying], time_delta, InsightType.Price, InsightDirection.Down, confidence=1-c) )
        
        return insights
import itertools
from datetime import timedelta
from collections import defaultdict
from Utilities.Unsorted import *
from Risk.CompositeRiskManagementModel import CompositeRiskManagementModel
from Risk.TrailingStopRiskManagementModel import TrailingStopRiskManagementModel
from Risk.MaximumDrawdownPercentPerSecurity import MaximumDrawdownPercentPerSecurity
from Portfolio.MeanVarianceOptimizationPortfolioConstructionModel import MeanVarianceOptimizationPortfolioConstructionModel

class StandardAlgorithm(QCAlgorithmFramework):
    Timeframes = []
    
    
    def __init__(self):
        # TODO: Remove self.Alphas?
        self.Alphas = []
        self.SetBroker()
        self.SetRisk()
        self.SetTiming()
        self.SetSecurityInitializer(self.SetSecurities)
    
    
    # Each alpha model is registered once per consolidated resolution
    def SetAlphas(self, *alpha_models):
        for alpha_model in alpha_models:
            alpha = alpha_model(self)
            self.AddChart(alpha.Chart)
            self.Alphas.append(alpha)
        
        self.SetWarmUp(max(map(lambda alpha: alpha.Indicator.Warmup, self.Alphas)))
        self.SetAlpha(CompositeAlphaModel(*self.Alphas))
    
    
    def SetBroker(self):
        self.SetBrokerageModel(BrokerageName.InteractiveBrokersBrokerage, AccountType.Margin)
    
    
    def SetSecurities(self, security):
        security.SetDataNormalizationMode(DataNormalizationMode.Raw)
    
    
    def OnSecuritiesChanged(self, changes):
        for consolidator in self.Consolidators:
            for security in changes.AddedSecurities:
                # self.SubscriptionManager.AddConsolidator(security.Symbol.Value, consolidator)
                pass
            for security in changes.RemovedSecurities:
                self.SubscriptionManager.RemoveConsolidator(security.Symbol.Value, consolidator)
    
    
    def SetRisk(self):
        self.SetCash(20000)
        trailing_stop = 0.10
        max_drawdown = 0.05
        self.SetRiskManagement(CompositeRiskManagementModel(TrailingStopRiskManagementModel(trailing_stop), MaximumDrawdownPercentPerSecurity(max_drawdown)))
        self.SetPortfolioConstruction(MeanVarianceOptimizationPortfolioConstructionModel())
        self.Allocate = 0.25


    def SetTiming(self):
        self.SetStartDate(2019, 2, 11)
        self.SetEndDate(2019, 2, 15)
        self.SetTimeZone(TimeZones.Toronto)
        self.UniverseSettings.Resolution = Resolution.Minute
        self.Consolidators = map(lambda t: TradeBarConsolidator(timedelta(minutes=t)), self.Timeframes)
def ConsolidatedSymbolLabel(symbol, consolidator):
    return "{} ({})".format(symbol, consolidator.Period)
from abc import ABC, abstractmethod


class StandardIndicator:
    Name = "Null Indicator"
    
    
    def __init__(self):
        self.Value = 0
        self.IsReady = False
    
    
    @abstractmethod
    def FillWindows(self, *args, value=0):
        for window in args:
            for i in range(window.Size):
                window.Add(value)