public QuadraticLinearRegression(double[] XDATA, double[] YDATA)
{
try
{
if ((XDATA.Length < 2 || YDATA.Length < 2) || (XDATA.Length != YDATA.Length))
{
throw new ArgumentOutOfRangeException("Exponential fit can't work with less then 2 points or unequal matrices");
}
if (XDATA.Length != YDATA.Length)
{
throw new ArgumentException("Arrays of unequal size were passed to the quadratic fitter");
}
//deep copy the data to protect its integrity
y = YDATA.ToArray();
y = (from b in y select Math.Log(b)).ToArray();
x = XDATA.ToArray();
DoubleArray dx = DoubleArray.From(XDATA);
DoubleArray dy = DoubleArray.From(y);
var x2 = dx.ElementMultiply(dx);
var xmat = DoubleArray.VertStack(dx, x2);
xmat = xmat.T;
r = new Regress(dy, xmat);
this.name = "ExpDecreasing";
pParameters = r.Beta;
this.SuccessfulFit = true;
}
catch
{
this.SuccessfulFit = false;
}
}
public static void Run()
{
int yearsToAggregate = 300;
bool expectedUtility = false;
double[] prtps = { 0.001, 0.01, 0.03 };
double[] etas = { 1.0, 1.5, 2.0 };
WelfareType[] swfs = { WelfareType.Global, WelfareType.Utilitarian, WelfareType.Regional, WelfareType.Tol, WelfareType.Pearce };
var stopwatch = new Stopwatch();
stopwatch.Start();
Console.WriteLine("Computing base CpC");
var baseCpC = Compute2010CpC();
Console.WriteLine("Doing a run with monteCarloRuns={0}, yearsToAggregate={1}",
monteCarloRuns,
yearsToAggregate);
var runConfigs = (from prtp in prtps
from eta in etas
from swf in swfs
select new RunConfig()
{
Prtp = prtp,
Eta = eta,
SWF = swf
}).ToArray();
string[] regions = new string[16];
var parameterDefinition = new Parameters();
parameterDefinition.ReadExcelFile(@"Data\Parameter - base.xlsm");
// Run model once to prep for multi tasking
{
var m = new Esmf.Model.ModelTyped <FundWorkflow>();
m.Run(parameterDefinition.GetBestGuess());
}
var relevantKeys = new List <ParameterElementKey>();
foreach (var p in parameterDefinition.GetElements())
{
if (p is ParameterElement <double> )
{
if (!(p is ParameterElementConstant <double>))
{
relevantKeys.Add(p.Key);
}
}
}
var yValues = new DoubleArray[runConfigs.Length, 16];
var correlations = new Dictionary <ParameterElementKey, double> [runConfigs.Length, 16];
var regressions = new Dictionary <ParameterElementKey, double> [runConfigs.Length, 16];
var regressionsConfIntervLow = new Dictionary <ParameterElementKey, double> [runConfigs.Length, 16];
var regressionsConfIntervHigh = new Dictionary <ParameterElementKey, double> [runConfigs.Length, 16];
var regressionsFStat = new DoubleArray(runConfigs.Length, 16);
var regressionsPVal = new DoubleArray(runConfigs.Length, 16);
var regressionsRsq = new DoubleArray(runConfigs.Length, 16);
var xValues = new DoubleArray(monteCarloRuns, relevantKeys.Count);
var standardizedXValues = new DoubleArray(monteCarloRuns, relevantKeys.Count);
var standardizedYValues = new DoubleArray[runConfigs.Length, 16];
for (int i = 0; i < runConfigs.Length; i++)
{
for (int l = 0; l < 16; l++)
{
yValues[i, l] = new DoubleArray(monteCarloRuns, 1);
standardizedYValues[i, l] = new DoubleArray(monteCarloRuns);
correlations[i, l] = new Dictionary <ParameterElementKey, double>();
regressions[i, l] = new Dictionary <ParameterElementKey, double>();
regressionsConfIntervHigh[i, l] = new Dictionary <ParameterElementKey, double>();
regressionsConfIntervLow[i, l] = new Dictionary <ParameterElementKey, double>();
}
}
var xMeans = new DoubleArray(relevantKeys.Count);
var xStd = new DoubleArray(relevantKeys.Count);
var rand = new MersenneTwister();
int currentRun = 0;
{
var m = new MarginalDamage2()
{
EmissionYear = Timestep.FromYear(2010),
Gas = MarginalGas.C,
Parameters = parameterDefinition.GetBestGuess(),
YearsToAggregate = yearsToAggregate,
GlobalCpCAtBase = baseCpC.Item1,
ExpectedUtilityMode = expectedUtility,
AdditionalInitMethod = (Esmf.Model.Model fw) =>
{
fw["scenariouncertainty"].Parameters["timeofuncertaintystart"].SetValue(Timestep.FromYear(2010));
}
};
for (int i = 0; i < 16; i++)
{
m.RegionalCpCAtBase[i] = baseCpC.Item2[i];
}
for (int i = 0; i < runConfigs.Length; i++)
{
m.SWFs.Add(new WelfareSpec(runConfigs[i].SWF, runConfigs[i].Prtp, runConfigs[i].Eta));
}
m.Start();
}
Parallel.ForEach(
parameterDefinition.GetRandom(rand, monteCarloRuns),
() =>
{
Thread.CurrentThread.CurrentCulture = CultureInfo.InvariantCulture;
Thread.CurrentThread.Priority = ThreadPriority.BelowNormal;
return(0);
},
(pv, pls, dummy) =>
{
int tempCurrentCount = Interlocked.Increment(ref currentRun);
Console.Write("\rRun {0} ", tempCurrentCount);
var m = new MarginalDamage2()
{
EmissionYear = Timestep.FromYear(2010),
Gas = MarginalGas.C,
Parameters = pv,
YearsToAggregate = yearsToAggregate,
GlobalCpCAtBase = baseCpC.Item1,
ExpectedUtilityMode = false,
AdditionalInitMethod = (Esmf.Model.Model fw) =>
{
fw["scenariouncertainty"].Parameters["timeofuncertaintystart"].SetValue(Timestep.FromYear(2010));
}
};
for (int i = 0; i < 16; i++)
{
m.RegionalCpCAtBase[i] = baseCpC.Item2[i];
}
for (int i = 0; i < runConfigs.Length; i++)
{
m.SWFs.Add(new WelfareSpec(runConfigs[i].SWF, runConfigs[i].Prtp, runConfigs[i].Eta));
}
m.Start();
var dimensions = m.Result1.Dimensions;
for (int i = 0; i < runConfigs.Length; i++)
{
foreach (var r in dimensions.GetValues <Region>())
{
regions[r.Index] = r.ToString();
yValues[i, r.Index][(int)pv.RunId - 1] = m.SCCs[i][r];
}
}
for (int l = 0; l < relevantKeys.Count; l++)
{
var p = relevantKeys[l];
double val = ((ParameterValueElement <double>)pv.GetElementByKey(p)).Value;
xValues[pv.RunId.Value - 1, l] = val;
}
return(0);
},
(dummy) => { });
Console.WriteLine();
Console.WriteLine();
if (!Directory.Exists("Output"))
{
Directory.CreateDirectory("Output");
}
Console.WriteLine("Write summary");
// Output summary
using (var f = File.CreateText(@"Output\summary.csv"))
{
f.WriteLine("Prtp;Eta;Swf;SccRegion;Mean;Variance;StdDev;StandardError");
for (int i = 0; i < runConfigs.Length; i++)
{
for (int l = 0; l < 16; l++)
{
double standardError = yValues[i, l].Std() / Math.Sqrt(monteCarloRuns);
f.WriteLine("{0};{1};{2:f3};{3:f3};{4:f3};{5:f3}",
runConfigs[i],
regions[l],
yValues[i, l].Mean(),
yValues[i, l].Var(),
yValues[i, l].Std(),
standardError);
}
}
}
Console.WriteLine("Compute correlations");
// Compute correlations
for (int k = 0; k < relevantKeys.Count; k++)
{
var xSlice = xValues.GetCol(k);
for (int i = 0; i < runConfigs.Length; i++)
{
for (int l = 0; l < 16; l++)
{
// Compute correlation between parameter i and SCC
var pearson = new Pearson(yValues[i, l], xSlice);
double corr = (double)pearson.Rho;
correlations[i, l].Add(relevantKeys[k], corr);
}
}
}
Console.WriteLine("Compute regressions");
// Standardize everything
for (int k = 0; k < relevantKeys.Count; k++)
{
var xSlice = xValues.GetCol(k);
xMeans[k] = xSlice.Mean();
xStd[k] = xSlice.Std();
}
for (int k = 0; k < monteCarloRuns; k++)
{
for (int i = 0; i < runConfigs.Length; i++)
{
for (int l = 0; l < 16; l++)
{
standardizedYValues[i, l][k] = (yValues[i, l][k] - yValues[i, l].Mean()) / yValues[i, l].Std();
}
}
for (int l = 0; l < relevantKeys.Count; l++)
{
standardizedXValues[k, l] = (xValues[k, l] - xMeans[l]) / xStd[l];
}
}
// Compute regression
for (int i = 0; i < runConfigs.Length; i++)
{
for (int l = 0; l < 16; l++)
{
var regress = new Regress(standardizedYValues[i, l], standardizedXValues, 0.1);
regressionsFStat[i, l] = regress.FStat;
regressionsPVal[i, l] = regress.PVal;
regressionsRsq[i, l] = regress.Rsq;
for (int k = 0; k < relevantKeys.Count; k++)
{
regressions[i, l].Add(relevantKeys[k], regress.Beta[k + 1]);
regressionsConfIntervLow[i, l].Add(relevantKeys[k], regress.BetaInt[k + 1, 0]);
regressionsConfIntervHigh[i, l].Add(relevantKeys[k], regress.BetaInt[k + 1, 1]);
}
}
}
Console.WriteLine("Write regression summary");
// Write regression summaries
using (var f = File.CreateText(@"Output\regression summary.csv"))
{
f.WriteLine("Prtp;Eta;Swf;SccRegion;FStat;PVal;Rsq");
for (int i = 0; i < runConfigs.Length; i++)
{
for (int l = 0; l < 16; l++)
{
f.WriteLine("{0};{1};{2:f15};{3:f15};{4:f15}", runConfigs[i], regions[l], regressionsFStat[i, l], regressionsPVal[i, l], regressionsRsq[i, l]);
}
}
}
Console.WriteLine("Write correlation");
// Write correlation
using (var f = File.CreateText(@"Output\correlation.csv"))
{
f.WriteLine("Prtp;Eta;Swf;SccRegion;Name;Correlation;RegressCoefficient;RegressConfIntLower;RegressConfIntUpper");
for (int i = 0; i < runConfigs.Length; i++)
{
for (int l = 0; l < 16; l++)
{
foreach (var key in relevantKeys)
{
string s = key.Name;
if (key is ParameterElementKey1Dimensional)
{
s += "-" + regions[((ParameterElementKey1Dimensional)key).D1];
}
else if (key is ParameterElementKey2Dimensional)
{
s += "-" + regions[((ParameterElementKey2Dimensional)key).D1] + "-" + regions[((ParameterElementKey2Dimensional)key).D2];
}
f.WriteLine("{0};{1};\"{2}\";{3:f10};{4:f10};{5:f10};{6:f10}", runConfigs[i], regions[l], s, correlations[i, l][key], regressions[i, l][key], regressionsConfIntervLow[i, l][key], regressionsConfIntervHigh[i, l][key]);
}
}
}
}
Console.WriteLine("Write SCC values");
using (var f = File.CreateText(@"Output\scc values.csv"))
{
f.WriteLine("Prtp;Eta;Swf;SccRegion;RundId;Scc");
for (int i = 0; i < runConfigs.Length; i++)
{
for (int l = 0; l < 16; l++)
{
for (int k = 0; k < monteCarloRuns; k++)
{
double standardError = yValues[i, l].Std() / Math.Sqrt(monteCarloRuns);
f.WriteLine("{0};{1};{2};{3:f14}",
runConfigs[i],
regions[l],
k,
yValues[i, l][k]);
}
}
}
}
stopwatch.Stop();
Console.WriteLine(stopwatch.Elapsed);
}
// 根据单调性对栅格打分
private Result monotoneLoc()
{
int agMaxX = 0, agMaxY = 0, agMinX = 0, agMinY = 0, agZ = 0;
GridHelper.getInstance().XYZToAccGrid(leftBound, downBound, 0, ref agMinX, ref agMinY, ref agZ);
GridHelper.getInstance().XYZToAccGrid(rightBound, upBound, 0, ref agMaxX, ref agMaxY, ref agZ);
Dictionary <string, int> dicMon = new Dictionary <string, int>();
Regress regress1 = new Regress();
for (int div = 0; div < roadDivide.Count - 1; div++)
{
List <double> pwrDbmSub = new List <double>();
List <double> x = new List <double>();
x.Add(0);
for (int j = roadDivide[div]; j < roadDivide[div + 1]; j++)
{
pwrDbmSub.Add(pwrDbm[j]);
x.Add(j - roadDivide[div] + 1);
}
#region 提取信号强度突变点
Divide divide = new Divide(100, 0.8, ref pwrDbmSub);
divide.run();
#endregion
#region 对每段进行线性拟合
for (int i = 0; i < divide.posV.Count - 1; i++)
{
int id1 = divide.posV[i];
int id2 = divide.posV[i + 1] - 1;
if (id2 < 0)
{
id2 = 0;
}
double a2 = 0, b2 = 0, err2 = 0;
regress1.CalcRegress(ref x, divide.m, id1, id2, out a2, out b2, out err2); // y = ax + b
double x1, y1, x2, y2; // (x1, y1) 单调上升的尽头
if (a2 < 0)
{
x1 = gx[id1 + roadDivide[div]];
y1 = gy[id1 + roadDivide[div]];
x2 = gx[id2 + roadDivide[div]];
y2 = gy[id2 + roadDivide[div]];
}
else
{
x1 = gx[id2 + roadDivide[div]];
y1 = gy[id2 + roadDivide[div]];
x2 = gx[id1 + roadDivide[div]];
y2 = gy[id1 + roadDivide[div]];
}
double k = (x1 - x2) / (y2 - y1);
double b = y1 - (x1 - x2) / (y2 - y1) * x1;
double y2Tmp = k * x2 + b;
if (y2 < y2Tmp)
{
for (int tx = agMinX + 1; tx < agMaxX - 1; tx++)
{
for (int ty = agMinY + 1; ty < agMaxY - 1; ty++)
{
double cx = 0, cy = 0, cz = 0;
GridHelper.getInstance().AccGridToXYZ(tx, ty, 0, ref cx, ref cy, ref cz);
if (cy > k * cx + b)
{
string key = string.Format("{0},{1}", tx, ty);
if (dicMon.Keys.Contains(key))
{
dicMon[key]++;
}
else
{
dicMon[key] = 0;
}
}
}
}
}
else
{
for (int tx = agMinX + 1; tx < agMaxX - 1; tx++)
{
for (int ty = agMinY + 1; ty < agMaxY - 1; ty++)
{
double cx = 0, cy = 0, cz = 0;
GridHelper.getInstance().AccGridToXYZ(tx, ty, 0, ref cx, ref cy, ref cz);
if (cy < k * cx + b)
{
string key = string.Format("{0},{1}", tx, ty);
if (dicMon.Keys.Contains(key))
{
dicMon[key]++;
}
else
{
dicMon[key] = 0;
}
}
}
}
}
}
#endregion
}
// 挑选分数最高的格子
if (dicMon.Count == 0)
{
return(new Result(false, "单调性未推出合理结果"));
}
var maxKeyMon = (from d in dicMon orderby d.Value descending select d.Key).First();
var allMaxKeyMon = (from d in dicMon where d.Value == dicMon[maxKeyMon] select d.Key).ToArray();
double X = 0, Y = 0, Z = 0;
int cnt = 0;
if (allMaxKeyMon.Count() > 5)
{
foreach (string a in allMaxKeyMon) // 有多个分数最高的格子
{
string[] num = a.Split(',');
GridHelper.getInstance().AccGridToXYZ(Convert.ToInt32(num[0]), Convert.ToInt32(num[1]), 0, ref X, ref Y, ref Z);
Pt3D pt = new Pt3D(X, Y, 0);
if (inBound(ref pt))
{
MonotoneResult.Add(pt);
}
}
}
else
{
var dicSort = from objDic in dicMon orderby objDic.Value descending select objDic;
foreach (KeyValuePair <string, int> kvp in dicSort)
{
string[] num = kvp.Key.Split(',');
GridHelper.getInstance().AccGridToXYZ(Convert.ToInt32(num[0]), Convert.ToInt32(num[1]), 0, ref X, ref Y, ref Z);
++cnt;
MonotoneResult.Add(new Pt3D(X, Y, 0));
if (cnt > 5)
{
break;
}
}
}
return(new Result(true));
}
public void LoadData()
{
DataLoad.Run();
var range = _meltDataMdb
.FindMeltRange(MeltNumber)
.Select(x => x.ToDouble())
.ToArray();
Tube.Чугун.C = range[2];
Tube.Чугун.Si = range[3];
Tube.Чугун.Mn = range[4];
Tube.Чугун.P = range[5];
Tube.Чугун.S = range[6];
Tube.Лом.C = range[7];
Tube.Лом.Si = range[8];
Tube.Лом.Mn = range[9];
Tube.Лом.P = range[10];
Tube.Лом.S = range[11];
Tube.Чугун.GEstimated = range[12];
Tube.Лом.GEstimated = range[13];
Tube.ОставленныйШлак.G = range[14] * 1000;
Tube.МиксерныйШлак.G = range[15] * 1000;
Tube.Шлак.GEnd = range[16] * 1000;
Tube.Сталь.GYield = range[17] * 1000;
Tube.Сталь.C = range[18];
Tube.Сталь.Si = range[19];
Tube.Сталь.Mn = range[20];
Tube.Сталь.P = range[21];
Tube.Сталь.S = range[22];
Tube.Шлак.FeOEnd = range[23];
Tube.Шлак.MnOEnd = range[24];
Tube.Шлак.P2O5End = range[25];
Tube.Шлак.BEnd = range[26];
Tube.Известь.G = range[27] * 1000;
Tube.Известняк.G = range[28] * 1000;
Tube.Доломит.G = range[29] * 1000;
Tube.ВлажныйДоломит.G = range[30] * 1000;
Tube.Имф.G = range[31] * 1000;
Tube.Песок.G = range[32] * 1000;
Tube.Кокс.G = range[33] * 1000;
Tube.Окатыши.G = range[34] * 1000;
Tube.уда.G = range[35] * 1000;
Tube.Окалина.G = range[36] * 1000;
Tube.Агломерат.G = range[37] * 1000;
Tube.Шпат.G = range[38] * 1000;
Tube.Дутье.V = range[39];
Params.AirTemp = Convert.ToInt32(range[40]) + 273;
Tube.Чугун.T = range[41] + 273;
Tube.Сталь.T = range[42] + 273;
Tube.Дутье.VarBlow[0] = range[43];
Tube.Лом.DolyaLegkovesa = range[51];
Regress.Load();
}
