| 1 | ////////////////////////////////////////////////////////////////////////////// |
|---|
| 2 | // FILE : |
|---|
| 3 | // PURPOSE : |
|---|
| 4 | ////////////////////////////////////////////////////////////////////////////// |
|---|
| 5 | |
|---|
| 6 | |
|---|
| 7 | //_.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-. |
|---|
| 8 | NameBlock Logit = |
|---|
| 9 | //_.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-. |
|---|
| 10 | [[ |
|---|
| 11 | Real _.ctInfo = 0; |
|---|
| 12 | Real _.dEpsilon = DiffDist; |
|---|
| 13 | Real _.maxIter = MaxIter; |
|---|
| 14 | Real _.tolerance = 10^(-8); |
|---|
| 15 | Real _.tolerance.Rec = 10^(-4); |
|---|
| 16 | Real _.probability.Init = 1/2; |
|---|
| 17 | Real _.num.Step = 8; |
|---|
| 18 | |
|---|
| 19 | Matrix _Error(Matrix y, Matrix p) |
|---|
| 20 | { |
|---|
| 21 | y-p //p = Probability(X, B) |
|---|
| 22 | }; |
|---|
| 23 | |
|---|
| 24 | Matrix _Gradient(Matrix X, Matrix error) |
|---|
| 25 | { |
|---|
| 26 | Tra(Tra(error)*X) //error = y - Probability(X, B) |
|---|
| 27 | }; |
|---|
| 28 | |
|---|
| 29 | Matrix _Hessian(VMatrix vX, Matrix p) |
|---|
| 30 | { |
|---|
| 31 | VMatrix W = Mat2VMat(p$*RSum(-p,1)); |
|---|
| 32 | VMatrix V = Eye(Rows(p), Rows(p), 0, W); |
|---|
| 33 | VMat2Mat(-Tra(V*vX)*vX) |
|---|
| 34 | }; |
|---|
| 35 | |
|---|
| 36 | Matrix _Probability(Matrix X, Matrix B, Matrix LogitGroupProb) |
|---|
| 37 | { |
|---|
| 38 | Matrix p = RPow(RSum(Exp(-X*B-LogitGroupProb), 1), -1); |
|---|
| 39 | p |
|---|
| 40 | }; |
|---|
| 41 | |
|---|
| 42 | Matrix _MLnLikelyhood(Matrix y, Matrix p) |
|---|
| 43 | { |
|---|
| 44 | y$*Log(p)+RSum(-y,1)$*Log(RSum(-p,1)) |
|---|
| 45 | }; |
|---|
| 46 | |
|---|
| 47 | Matrix _MLikelyhood(Matrix mLnLikelyhood) |
|---|
| 48 | { |
|---|
| 49 | Exp(mLnLikelyhood) |
|---|
| 50 | }; |
|---|
| 51 | |
|---|
| 52 | Set Estimate.MaxLikelyhood |
|---|
| 53 | ( |
|---|
| 54 | Matrix y, // Matriz y de variable salida binaria |
|---|
| 55 | Matrix XIni, // Matriz x de variables entrada |
|---|
| 56 | Matrix B0Ini, // Matriz de parametros iniciales |
|---|
| 57 | Anything ctInfo, // Un real 0 sin constante, |
|---|
| 58 | // Un real 1 estima constante, |
|---|
| 59 | // Una matriz |
|---|
| 60 | Real dEpsilon, // Diferencia de paso |
|---|
| 61 | Real maxIter, // Numero maximo de iteraciones |
|---|
| 62 | Real tolerance // Tolerancia al error |
|---|
| 63 | ) |
|---|
| 64 | { |
|---|
| 65 | Text grCtInfo = Grammar(ctInfo); |
|---|
| 66 | |
|---|
| 67 | Matrix X = Case |
|---|
| 68 | ( |
|---|
| 69 | /* |
|---|
| 70 | And(grCtInfo=="Real", EQ(ctInfo, 0)), XIni, |
|---|
| 71 | And(grCtInfo=="Real", EQ(ctInfo, 1)), |
|---|
| 72 | { |
|---|
| 73 | XIni|Rand(Rows(XIni), 1, 1, 1) |
|---|
| 74 | }, |
|---|
| 75 | */ |
|---|
| 76 | grCtInfo=="Matrix", XIni, |
|---|
| 77 | 1, XIni |
|---|
| 78 | ); |
|---|
| 79 | |
|---|
| 80 | Matrix IdCt = Case |
|---|
| 81 | ( |
|---|
| 82 | // And(grCtInfo=="Real", EQ(ctInfo, 0)), Rand(Rows(XIni), 1, 0, 0), |
|---|
| 83 | // And(grCtInfo=="Real", EQ(ctInfo, 1)), Rand(Rows(XIni), 1, 0, 0), |
|---|
| 84 | grCtInfo=="Matrix", ctInfo, |
|---|
| 85 | 1, Rand(Rows(XIni), 1, 0, 0) |
|---|
| 86 | ); |
|---|
| 87 | |
|---|
| 88 | Matrix B0 = Case |
|---|
| 89 | ( |
|---|
| 90 | grCtInfo=="Matrix", B0Ini, |
|---|
| 91 | 1, B0Ini |
|---|
| 92 | ); |
|---|
| 93 | |
|---|
| 94 | Real n = Columns(X); // Number of variables |
|---|
| 95 | Real N = Rows(X); // Number of data |
|---|
| 96 | |
|---|
| 97 | Text iniMsg = |
|---|
| 98 | "Model Logit Init ("+IntText(N)+"x"+IntText(n)+")"+Time; |
|---|
| 99 | Real WriteLn(iniMsg); |
|---|
| 100 | |
|---|
| 101 | VMatrix vX = Mat2VMat(X); |
|---|
| 102 | |
|---|
| 103 | Matrix zeroMat = Rand(N, 1, 0, 0); |
|---|
| 104 | Matrix oneMat = Rand(N, 1, 1, 1); |
|---|
| 105 | Matrix negInfMat = Rand(N, 1, -1/0, -1/0); |
|---|
| 106 | Matrix tolMat = Rand(N, 1, tolerance, tolerance); |
|---|
| 107 | |
|---|
| 108 | Real completeTime = 0; |
|---|
| 109 | Real difTimeMin = 0; |
|---|
| 110 | Real exit = 0; |
|---|
| 111 | Matrix B = Copy(B0); |
|---|
| 112 | Real iter = 0; |
|---|
| 113 | Real oldLnLikelyhood = 0; |
|---|
| 114 | |
|---|
| 115 | Set cycle = Empty; |
|---|
| 116 | |
|---|
| 117 | Real While(Not(exit), |
|---|
| 118 | { |
|---|
| 119 | Real time0 = Copy(Time); |
|---|
| 120 | |
|---|
| 121 | //Real CMsg::Trace::show(1, "Printing oldLnLikelyhood="<<oldLnLikelyhood); |
|---|
| 122 | Matrix p = _Probability(X, B, IdCt); |
|---|
| 123 | |
|---|
| 124 | //Real CMsg::Trace::show(1, "Printing p="<<p); |
|---|
| 125 | Matrix error = _Error(y, p); |
|---|
| 126 | |
|---|
| 127 | Matrix G = _Gradient(X, error); |
|---|
| 128 | Real time0.1 = Copy(Time); |
|---|
| 129 | Matrix H = _Hessian(vX, p); |
|---|
| 130 | Real time0.2 = Copy(Time); |
|---|
| 131 | Matrix dif = MinimumResidualsSolve(H, G); |
|---|
| 132 | |
|---|
| 133 | Real norm = MatFrobeniusNorm(G); |
|---|
| 134 | Real advance = MatFrobeniusNorm(dif); |
|---|
| 135 | Real maxAbsDif = MatMax(Abs(dif)); |
|---|
| 136 | |
|---|
| 137 | |
|---|
| 138 | Matrix mLnLikelyhood = _MLnLikelyhood(y, p); |
|---|
| 139 | Matrix mLLCorrect = IfMat(EQ(Abs(error), oneMat), negInfMat, |
|---|
| 140 | IfMat(LT(Abs(error), tolMat), zeroMat, mLnLikelyhood)); |
|---|
| 141 | |
|---|
| 142 | Real lnLikelyhood = MatSum(mLLCorrect); |
|---|
| 143 | //Real CMsg::Trace::show(1, "Printing dif"<<dif); |
|---|
| 144 | //Real CMsg::Trace::show(1, "Printing mLnLikelyhood"<<mLnLikelyhood); |
|---|
| 145 | //Real CMsg::Trace::show(1, "Printing mLLCorrect"<<mLLCorrect); |
|---|
| 146 | //Real CMsg::Trace::show(1, "Printing error"<<error); |
|---|
| 147 | |
|---|
| 148 | //Real CMsg::Trace::show(1, "Printing lnLikelyhood="<<lnLikelyhood); |
|---|
| 149 | |
|---|
| 150 | Real difTimeMin := (time0.2-time0.1)+difTimeMin; |
|---|
| 151 | |
|---|
| 152 | Real difTime = Copy(Time)-time0; |
|---|
| 153 | Real completeTime := completeTime+difTime; |
|---|
| 154 | Real exitTreatment = Case |
|---|
| 155 | ( |
|---|
| 156 | IsUnknown(advance), |
|---|
| 157 | { |
|---|
| 158 | Real CMsg::Trace::show(1, "Advance is unknown"); |
|---|
| 159 | Real exit:=1 |
|---|
| 160 | }, |
|---|
| 161 | GE(iter, maxIter), |
|---|
| 162 | { |
|---|
| 163 | Real CMsg::Trace::show(1, "MaxIter reached"); |
|---|
| 164 | Real exit:=1 |
|---|
| 165 | }, |
|---|
| 166 | LT(norm, dEpsilon), |
|---|
| 167 | { |
|---|
| 168 | Real CMsg::Trace::show(1, "Norm lower than "<<dEpsilon); |
|---|
| 169 | Real exit:=1 |
|---|
| 170 | }, |
|---|
| 171 | LT(maxAbsDif, tolerance), |
|---|
| 172 | { |
|---|
| 173 | Real CMsg::Trace::show(1, "MaxAbsDif lower than "<<tolerance); |
|---|
| 174 | Real exit:=1 |
|---|
| 175 | }, |
|---|
| 176 | LT(Abs(oldLnLikelyhood - lnLikelyhood), dEpsilon), |
|---|
| 177 | { |
|---|
| 178 | Real CMsg::Trace::show(1, |
|---|
| 179 | "LnLikelihood is stable: lnLikelyhood ="<<lnLikelyhood+ |
|---|
| 180 | " oldLnLikelyhood ="<<oldLnLikelyhood); |
|---|
| 181 | Real exit:=1 |
|---|
| 182 | }, |
|---|
| 183 | 1, |
|---|
| 184 | { |
|---|
| 185 | Text exitMsg = " Logit model iteration("+ |
|---|
| 186 | FormatReal(iter, "%0"+IntText(Floor(Log10(maxIter)+1))+".lf")+ |
|---|
| 187 | ")"+" "+ |
|---|
| 188 | " LogLikelyhood = "+FormatReal(lnLikelyhood, "%.4E")+" "+ |
|---|
| 189 | " MaxAbsDif = "+FormatReal(maxAbsDif, "%.4E")+" "+ |
|---|
| 190 | " Gradient Norm = "+FormatReal(norm, "%.4E")+" "+ |
|---|
| 191 | " Time = "+FormatReal(difTime, "%.4lf"); |
|---|
| 192 | Real CMsg::Trace::show(1, exitMsg); |
|---|
| 193 | Real iter := iter+1; |
|---|
| 194 | Real oldLnLikelyhood := Copy(lnLikelyhood); |
|---|
| 195 | Matrix B:=(Matrix B-dif); |
|---|
| 196 | Real 0 |
|---|
| 197 | } |
|---|
| 198 | ); |
|---|
| 199 | Real exitCycle = If(EQ(exitTreatment, 1), |
|---|
| 200 | { |
|---|
| 201 | Set cycle := SetOfAnything |
|---|
| 202 | (B, p, G, H, dif, norm, advance, |
|---|
| 203 | maxAbsDif, error, lnLikelyhood, oldLnLikelyhood, ctInfo, X, XIni, y, |
|---|
| 204 | mLLCorrect); |
|---|
| 205 | 1 |
|---|
| 206 | }, 0) |
|---|
| 207 | }); |
|---|
| 208 | Real time2 = Copy(Time); |
|---|
| 209 | /* |
|---|
| 210 | Matrix p = _Probability(X, B); |
|---|
| 211 | |
|---|
| 212 | Matrix error = _Error(y, p); |
|---|
| 213 | |
|---|
| 214 | Matrix G = _Gradient(X, error); |
|---|
| 215 | Matrix H = _Hessian(vX, p); |
|---|
| 216 | Matrix dif = MinimumResidualsSolve(H, G); |
|---|
| 217 | |
|---|
| 218 | Real norm = MatFrobeniusNorm(G); |
|---|
| 219 | Real advance = MatFrobeniusNorm(dif); |
|---|
| 220 | Real maxAbsDif = MatMax(Abs(dif)); |
|---|
| 221 | Matrix mLnLikelyhood = _MLnLikelyhood(y, p); |
|---|
| 222 | Real lnLikelyhood = MatSum(mLnLikelyhood); |
|---|
| 223 | */ |
|---|
| 224 | Real difTime2 = Copy(Time)-time2; |
|---|
| 225 | Real totalTime = completeTime+difTime2; |
|---|
| 226 | Text endMsg = |
|---|
| 227 | "Model Logit Ended. Time:"<<totalTime+" SplitTime:"<<difTimeMin+NL+NL; |
|---|
| 228 | Real WriteLn(endMsg); |
|---|
| 229 | |
|---|
| 230 | /* |
|---|
| 231 | SetOfAnything |
|---|
| 232 | (B, p, G, H, dif, norm, advance, |
|---|
| 233 | maxAbsDif, error, lnLikelyhood, oldLnLikelyhood) |
|---|
| 234 | */ |
|---|
| 235 | cycle |
|---|
| 236 | }; |
|---|
| 237 | |
|---|
| 238 | Set Estimate.MaxLikelyhood.Default(Matrix y, Matrix X) |
|---|
| 239 | { |
|---|
| 240 | Matrix B0Ini = Rand(Columns(X), 1, 0, 0); |
|---|
| 241 | Estimate.MaxLikelyhood |
|---|
| 242 | ( |
|---|
| 243 | y, // Matriz y de variable salida binaria |
|---|
| 244 | X, // Matriz x de variables entrada |
|---|
| 245 | B0Ini, |
|---|
| 246 | _.ctInfo, |
|---|
| 247 | _.dEpsilon, // Diferencia de paso |
|---|
| 248 | _.maxIter, // Numero maximo de iteraciones |
|---|
| 249 | _.tolerance // Tolerancia al error |
|---|
| 250 | ) |
|---|
| 251 | }; |
|---|
| 252 | Set Estimate.MaxLikelyhood.Constant(Matrix y, Matrix X, Matrix constant) |
|---|
| 253 | { |
|---|
| 254 | Matrix B0Ini = Rand(Columns(X), 1, 0, 0); |
|---|
| 255 | Estimate.MaxLikelyhood |
|---|
| 256 | ( |
|---|
| 257 | y, // Matriz y de variable salida binaria |
|---|
| 258 | X, // Matriz x de variables entrada |
|---|
| 259 | B0Ini, |
|---|
| 260 | constant, |
|---|
| 261 | _.dEpsilon, // Diferencia de paso |
|---|
| 262 | _.maxIter, // Numero maximo de iteraciones |
|---|
| 263 | _.tolerance // Tolerancia al error |
|---|
| 264 | ) |
|---|
| 265 | }; |
|---|
| 266 | Set Estimate.MaxLikelyhood.ProbB0 |
|---|
| 267 | ( |
|---|
| 268 | Matrix y, |
|---|
| 269 | Matrix X, |
|---|
| 270 | Matrix B0Ini, |
|---|
| 271 | Real prob |
|---|
| 272 | ) |
|---|
| 273 | { |
|---|
| 274 | Matrix probMat = Rand(Rows(y), 1, prob, prob); |
|---|
| 275 | Matrix constant = Log(probMat$/RSum(-probMat, 1)); |
|---|
| 276 | |
|---|
| 277 | Estimate.MaxLikelyhood |
|---|
| 278 | ( |
|---|
| 279 | y, // Matriz y de variable salida binaria |
|---|
| 280 | X, // Matriz x de variables entrada |
|---|
| 281 | B0Ini, |
|---|
| 282 | constant, |
|---|
| 283 | _.dEpsilon, // Diferencia de paso |
|---|
| 284 | _.maxIter, // Numero maximo de iteraciones |
|---|
| 285 | _.tolerance // Tolerancia al error |
|---|
| 286 | ) |
|---|
| 287 | }; |
|---|
| 288 | |
|---|
| 289 | Set Estimate.MaxLikelyhood.ProbRec |
|---|
| 290 | ( |
|---|
| 291 | Matrix y, |
|---|
| 292 | Matrix X, |
|---|
| 293 | Matrix B0Cur, |
|---|
| 294 | Real step, |
|---|
| 295 | Real tolerance, |
|---|
| 296 | Real probCur, |
|---|
| 297 | Real probEnd |
|---|
| 298 | ) |
|---|
| 299 | { |
|---|
| 300 | Text iniMsg = NL+NL+"Model Logit (p="<<probCur+") "+Time; |
|---|
| 301 | Real WriteLn(iniMsg); |
|---|
| 302 | |
|---|
| 303 | Set logitResult = Estimate.MaxLikelyhood.ProbB0(y, X, B0Cur, probCur); |
|---|
| 304 | If(LT(Abs(probCur - probEnd), tolerance), logitResult, |
|---|
| 305 | { |
|---|
| 306 | Real prob = probCur*step; |
|---|
| 307 | Matrix B0 = logitResult::B; |
|---|
| 308 | Estimate.MaxLikelyhood.ProbRec |
|---|
| 309 | (y, X, B0, step, tolerance, prob, probEnd) |
|---|
| 310 | }) |
|---|
| 311 | }; |
|---|
| 312 | |
|---|
| 313 | Set Estimate.MaxLikelyhood.Prob |
|---|
| 314 | ( |
|---|
| 315 | Matrix y, |
|---|
| 316 | Matrix X, |
|---|
| 317 | Real prob |
|---|
| 318 | ) |
|---|
| 319 | { |
|---|
| 320 | Real step = (prob/_.probability.Init)^(1/_.num.Step); |
|---|
| 321 | Matrix B0Ini = Rand(Columns(X), 1, 0, 0); |
|---|
| 322 | Estimate.MaxLikelyhood.ProbRec |
|---|
| 323 | (y, X, B0Ini, step, _.tolerance.Rec, _.probability.Init, prob) |
|---|
| 324 | }; |
|---|
| 325 | |
|---|
| 326 | Set Diagnosis |
|---|
| 327 | ( |
|---|
| 328 | Matrix y, |
|---|
| 329 | Matrix X, |
|---|
| 330 | Matrix B, |
|---|
| 331 | Matrix error, |
|---|
| 332 | Real lnLikelyhood, |
|---|
| 333 | Set names, |
|---|
| 334 | Matrix H //Hessian |
|---|
| 335 | ) |
|---|
| 336 | { |
|---|
| 337 | Text iniMsg = |
|---|
| 338 | "Model Logit Diagnosis. Init Time:"+Time; |
|---|
| 339 | Real WriteLn(iniMsg); |
|---|
| 340 | |
|---|
| 341 | Real N = Rows(X); // |
|---|
| 342 | Real n = Columns(X); // Parameter number |
|---|
| 343 | |
|---|
| 344 | Matrix FIM = -H; // Fisher Information Matrix |
|---|
| 345 | Matrix COV = SVDInverse(FIM); // Varianze Covarianza Parameter Matrix |
|---|
| 346 | |
|---|
| 347 | Set Parameters = For(1, Columns(X), Set(Real k) |
|---|
| 348 | { |
|---|
| 349 | Text name = names[k]; |
|---|
| 350 | Real value = MatDat(B, k, 1); |
|---|
| 351 | Real var = MatDat(COV, k, k); |
|---|
| 352 | Real std = SqRt(var); |
|---|
| 353 | Real tStudent = value/std; |
|---|
| 354 | Real refProb = 2*(1-DistT(Abs(tStudent), N-n-1)); |
|---|
| 355 | ParameterInf |
|---|
| 356 | ( |
|---|
| 357 | name, //Name |
|---|
| 358 | 0, //Factor |
|---|
| 359 | 0, //Order |
|---|
| 360 | value, //Value |
|---|
| 361 | std, //StDs |
|---|
| 362 | tStudent, //TStudent |
|---|
| 363 | refProb //RefuseProb |
|---|
| 364 | ) |
|---|
| 365 | }); |
|---|
| 366 | |
|---|
| 367 | Real VarTot = MatVar(y); |
|---|
| 368 | Real VarError = MatVar(error); |
|---|
| 369 | Real R2 = 1-VarError/VarTot; |
|---|
| 370 | |
|---|
| 371 | Real MaxProb = MatSum(y)/N; |
|---|
| 372 | Real lnLikelyhoodIntercep = |
|---|
| 373 | MatSum(_MLnLikelyhood(y, Rand(N, 1, MaxProb, MaxProb))); |
|---|
| 374 | |
|---|
| 375 | Real Nagelkerke.R2 = 1-(Exp((2/N)*(lnLikelyhoodIntercep-lnLikelyhood))); |
|---|
| 376 | Real Nagelkerke.R2Max = 1-Exp((2/N)*lnLikelyhoodIntercep); |
|---|
| 377 | Real Nagelkerke.R2MaxRescaled = Nagelkerke.R2/Nagelkerke.R2Max; |
|---|
| 378 | Text endMsg = |
|---|
| 379 | "Model Logit Diagnosis. End Time:"+Time+NL; |
|---|
| 380 | Real WriteLn(endMsg); |
|---|
| 381 | |
|---|
| 382 | SetOfAnything |
|---|
| 383 | ( |
|---|
| 384 | Parameters, |
|---|
| 385 | FIM, |
|---|
| 386 | COV, |
|---|
| 387 | MaxProb, |
|---|
| 388 | R2, |
|---|
| 389 | Nagelkerke.R2, |
|---|
| 390 | Nagelkerke.R2Max, |
|---|
| 391 | Nagelkerke.R2MaxRescaled, |
|---|
| 392 | lnLikelyhood, |
|---|
| 393 | lnLikelyhoodIntercep |
|---|
| 394 | ) |
|---|
| 395 | }; |
|---|
| 396 | |
|---|
| 397 | Set PreTesting(Matrix Y, Matrix X, Set varNames) |
|---|
| 398 | { |
|---|
| 399 | Text iniMsg = |
|---|
| 400 | "Model Logit PreTesting. Init Time:"+Time; |
|---|
| 401 | Real WriteLn(iniMsg); |
|---|
| 402 | |
|---|
| 403 | Real WriteLn(" Checking column stability..."+Time); |
|---|
| 404 | Matrix unkX = IsUnknown(X); |
|---|
| 405 | Matrix posInfX = IsPosInf(X); |
|---|
| 406 | Matrix negInfX = IsNegInf(X); |
|---|
| 407 | |
|---|
| 408 | Matrix unkY = IsUnknown(Y); |
|---|
| 409 | Matrix posInfY = IsPosInf(Y); |
|---|
| 410 | Matrix negInfY = IsNegInf(Y); |
|---|
| 411 | |
|---|
| 412 | Real isUnkX = MatSum(unkX); |
|---|
| 413 | Real isPosInfX = MatSum(posInfX); |
|---|
| 414 | Real isNegInfX = MatSum(negInfX); |
|---|
| 415 | |
|---|
| 416 | Real isUnkY = MatSum(unkY); |
|---|
| 417 | Real isPosInfY = MatSum(posInfY); |
|---|
| 418 | Real isNegInfY = MatSum(negInfY); |
|---|
| 419 | |
|---|
| 420 | Real n = Rows(Y); |
|---|
| 421 | Real balanced = MatSum(Y)/n; |
|---|
| 422 | |
|---|
| 423 | Real valid = |
|---|
| 424 | Not(Or(isUnkX, isPosInfX, isNegInfX, isUnkY, isPosInfY, isNegInfY)); |
|---|
| 425 | |
|---|
| 426 | Set checkValid = If(EQ(valid, 1), Empty, |
|---|
| 427 | { |
|---|
| 428 | Set data = For(1, Card(varNames), Set(Real k) |
|---|
| 429 | { |
|---|
| 430 | Text name = varNames[k]; |
|---|
| 431 | Real kUnkX = MatSum(SubCol(unkX, [[k]])); |
|---|
| 432 | Real kPosInfX = MatSum(SubCol(posInfX, [[k]])); |
|---|
| 433 | Real kNegInfX = MatSum(SubCol(negInfX, [[k]])); |
|---|
| 434 | SetOfAnything(name, kUnkX, kPosInfX, kNegInfX) |
|---|
| 435 | }); |
|---|
| 436 | Set header = SetOfText("VarName", "Unk", "PosInf", "NegInf"); |
|---|
| 437 | SetOfSet(header)<< |
|---|
| 438 | SetOfSet(SetOfAnything("Y", isUnkY, isPosInfY, isNegInfY))<< |
|---|
| 439 | data |
|---|
| 440 | }); |
|---|
| 441 | Real WriteLn(" Adjusting X matrix..."+Time); |
|---|
| 442 | |
|---|
| 443 | Matrix YPre = |
|---|
| 444 | If(isUnkY, IfMat(unkY, VMat2Mat(Eye(Rows(Y), 1, 0, 0)), Y), Y); |
|---|
| 445 | Matrix XPre = |
|---|
| 446 | If(isUnkX, IfMat(unkX, VMat2Mat(Eye(Rows(X), Columns(X), 0, 0)), X), X); |
|---|
| 447 | |
|---|
| 448 | VMatrix vXPre = Mat2VMat(XPre); |
|---|
| 449 | Set index = Range(1, Columns(X), 1); |
|---|
| 450 | // Real WriteLn(" Correlation Y|X matrix..."+Time); |
|---|
| 451 | // Matrix corVarX = Cor(Tra(YPre|XPre)); |
|---|
| 452 | |
|---|
| 453 | Real WriteLn(" X information column..."+Time); |
|---|
| 454 | Set ColInfo = For(1, Card(varNames), Set(Real k) |
|---|
| 455 | { |
|---|
| 456 | Text name = varNames[k]; |
|---|
| 457 | Matrix col = SubCol(XPre, [[k]]); |
|---|
| 458 | Real min = MatMin(col); |
|---|
| 459 | Real max = MatMax(col); |
|---|
| 460 | Real stds = MatStDs(col); |
|---|
| 461 | Real avr = MatAvr(col); |
|---|
| 462 | Matrix freq = Frequency(col, 100, min, max); |
|---|
| 463 | Matrix freq01 = Frequency(col$*Y, 100, min, max); |
|---|
| 464 | Matrix ratioDisc = freq01$/freq; |
|---|
| 465 | Real minValue = MatDat(freq01, 1, 2); |
|---|
| 466 | Real maxValue = MatDat(freq01, 100, 2); |
|---|
| 467 | Real maxRatio = MatDat(ratioDisc, 100, 2)/balanced; |
|---|
| 468 | Real dicotomicRatio = (minValue+maxValue)/n ; |
|---|
| 469 | |
|---|
| 470 | VMatrix y = SubCol(vXPre, [[k]]); |
|---|
| 471 | VMatrix x = SubCol(vXPre, index-[[k]]); |
|---|
| 472 | |
|---|
| 473 | Set linReg = LinReg::Get.GeneralInformation(y, x); |
|---|
| 474 | Real R2MultiColinearity = linReg::R2; |
|---|
| 475 | |
|---|
| 476 | SetOfAnything |
|---|
| 477 | ( |
|---|
| 478 | name, |
|---|
| 479 | min, |
|---|
| 480 | max, |
|---|
| 481 | stds, |
|---|
| 482 | avr, |
|---|
| 483 | freq, |
|---|
| 484 | freq01, |
|---|
| 485 | ratioDisc, |
|---|
| 486 | maxRatio, |
|---|
| 487 | maxValue, |
|---|
| 488 | dicotomicRatio, |
|---|
| 489 | R2MultiColinearity |
|---|
| 490 | ) |
|---|
| 491 | }); |
|---|
| 492 | |
|---|
| 493 | Text endMsg = |
|---|
| 494 | "Model Logit PreTesting. End Time:"+Time+NL; |
|---|
| 495 | Real WriteLn(endMsg); |
|---|
| 496 | |
|---|
| 497 | SetOfAnything |
|---|
| 498 | (unkX, posInfX, negInfX, unkY, posInfY, negInfY, valid, checkValid, |
|---|
| 499 | balanced, /*corVarX,*/ |
|---|
| 500 | ColInfo, YPre, XPre) |
|---|
| 501 | } |
|---|
| 502 | ]]; |
|---|
| 503 | //_.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-._.-. |
|---|
| 504 | |
|---|