[5571] | 1 | #region License Information
|
---|
| 2 | /* HeuristicLab
|
---|
[17180] | 3 | * Copyright (C) Heuristic and Evolutionary Algorithms Laboratory (HEAL)
|
---|
[5571] | 4 | *
|
---|
| 5 | * This file is part of HeuristicLab.
|
---|
| 6 | *
|
---|
| 7 | * HeuristicLab is free software: you can redistribute it and/or modify
|
---|
| 8 | * it under the terms of the GNU General Public License as published by
|
---|
| 9 | * the Free Software Foundation, either version 3 of the License, or
|
---|
| 10 | * (at your option) any later version.
|
---|
| 11 | *
|
---|
| 12 | * HeuristicLab is distributed in the hope that it will be useful,
|
---|
| 13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
---|
| 14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
---|
| 15 | * GNU General Public License for more details.
|
---|
| 16 | *
|
---|
| 17 | * You should have received a copy of the GNU General Public License
|
---|
| 18 | * along with HeuristicLab. If not, see <http://www.gnu.org/licenses/>.
|
---|
| 19 | */
|
---|
| 20 | #endregion
|
---|
| 21 |
|
---|
| 22 | using System;
|
---|
| 23 | using System.Collections.Generic;
|
---|
[17573] | 24 | using System.Linq;
|
---|
[5571] | 25 | using HeuristicLab.Common;
|
---|
| 26 | using HeuristicLab.Core;
|
---|
[6740] | 27 | using HeuristicLab.Data;
|
---|
[5571] | 28 | using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
|
---|
[6740] | 29 | using HeuristicLab.Parameters;
|
---|
[16565] | 30 | using HEAL.Attic;
|
---|
[17721] | 31 | using MathNet.Numerics;
|
---|
[17448] | 32 | using MathNet.Numerics.Statistics;
|
---|
[17455] | 33 | using DoubleVector = MathNet.Numerics.LinearAlgebra.Vector<double>;
|
---|
| 34 |
|
---|
[5571] | 35 | namespace HeuristicLab.Problems.DataAnalysis.Symbolic {
|
---|
[17455] | 36 | [StorableType("DE68A1D9-5AFC-4DDD-AB62-29F3B8FC28E0")]
|
---|
| 37 | [Item("SymbolicDataAnalysisExpressionTreeVectorInterpreter", "Interpreter for symbolic expression trees including vector arithmetic.")]
|
---|
| 38 | public class SymbolicDataAnalysisExpressionTreeVectorInterpreter : ParameterizedNamedItem, ISymbolicDataAnalysisExpressionTreeInterpreter {
|
---|
[17467] | 39 | [StorableType("2612504E-AD5F-4AE2-B60E-98A5AB59E164")]
|
---|
| 40 | public enum Aggregation {
|
---|
| 41 | Mean,
|
---|
| 42 | Median,
|
---|
| 43 | Sum,
|
---|
[17573] | 44 | First,
|
---|
[17721] | 45 | L1Norm,
|
---|
| 46 | L2Norm,
|
---|
[17467] | 47 | NaN,
|
---|
| 48 | Exception
|
---|
| 49 | }
|
---|
[17721] | 50 | public static double Aggregate(Aggregation aggregation, DoubleVector vector) {
|
---|
| 51 | switch (aggregation) {
|
---|
| 52 | case Aggregation.Mean: return Statistics.Mean(vector);
|
---|
| 53 | case Aggregation.Median: return Statistics.Median(vector);
|
---|
| 54 | case Aggregation.Sum: return vector.Sum();
|
---|
| 55 | case Aggregation.First: return vector.First();
|
---|
| 56 | case Aggregation.L1Norm: return vector.L1Norm();
|
---|
| 57 | case Aggregation.L2Norm: return vector.L2Norm();
|
---|
| 58 | case Aggregation.NaN: return double.NaN;
|
---|
| 59 | case Aggregation.Exception: throw new InvalidOperationException("Result of the tree is not a scalar.");
|
---|
| 60 | default: throw new ArgumentOutOfRangeException(nameof(aggregation), aggregation, null);
|
---|
| 61 | }
|
---|
| 62 | }
|
---|
[17455] | 63 |
|
---|
[17721] | 64 | [StorableType("73DCBB45-916F-4139-8ADC-57BA610A1B66")]
|
---|
| 65 | public enum VectorLengthStrategy {
|
---|
| 66 | ExceptionIfDifferent,
|
---|
| 67 | FillShorterWithNaN,
|
---|
| 68 | FillShorterWithNeutralElement,
|
---|
| 69 | CutLonger,
|
---|
| 70 | ResampleToLonger,
|
---|
| 71 | ResampleToShorter,
|
---|
| 72 | CycleShorter
|
---|
| 73 | }
|
---|
| 74 |
|
---|
| 75 | #region Implementation VectorLengthStrategy
|
---|
| 76 | public static (DoubleVector, DoubleVector) ExceptionIfDifferent(DoubleVector lhs, DoubleVector rhs) {
|
---|
| 77 | if (lhs.Count != rhs.Count)
|
---|
| 78 | throw new InvalidOperationException($"Vector Lengths incompatible ({lhs.Count} vs. {rhs.Count}");
|
---|
| 79 | return (lhs, rhs);
|
---|
| 80 | }
|
---|
| 81 |
|
---|
| 82 | public static (DoubleVector, DoubleVector) FillShorter(DoubleVector lhs, DoubleVector rhs, double fillElement) {
|
---|
| 83 | var targetLength = Math.Max(lhs.Count, rhs.Count);
|
---|
| 84 |
|
---|
| 85 | DoubleVector PadVector(DoubleVector v) {
|
---|
| 86 | if (v.Count == targetLength) return v;
|
---|
| 87 | var p = DoubleVector.Build.Dense(targetLength, fillElement);
|
---|
| 88 | v.CopySubVectorTo(p, 0, 0, v.Count);
|
---|
| 89 | return p;
|
---|
| 90 | }
|
---|
| 91 |
|
---|
| 92 | return (PadVector(lhs), PadVector(rhs));
|
---|
| 93 | }
|
---|
| 94 |
|
---|
| 95 | public static (DoubleVector, DoubleVector) CutLonger(DoubleVector lhs, DoubleVector rhs) {
|
---|
| 96 | var targetLength = Math.Min(lhs.Count, rhs.Count);
|
---|
| 97 |
|
---|
| 98 | DoubleVector CutVector(DoubleVector v) {
|
---|
| 99 | if (v.Count == targetLength) return v;
|
---|
| 100 | return v.SubVector(0, targetLength);
|
---|
| 101 | }
|
---|
| 102 |
|
---|
| 103 | return (CutVector(lhs), CutVector(rhs));
|
---|
| 104 | }
|
---|
| 105 |
|
---|
| 106 | private static DoubleVector ResampleToLength(DoubleVector v, int targetLength) {
|
---|
| 107 | if (v.Count == targetLength) return v;
|
---|
| 108 |
|
---|
| 109 | var indices = Enumerable.Range(0, v.Count).Select(x => (double)x);
|
---|
| 110 | var interpolation = Interpolate.Linear(indices, v);
|
---|
| 111 |
|
---|
| 112 | var resampledIndices = Enumerable.Range(0, targetLength).Select(i => (double)i / targetLength * v.Count);
|
---|
| 113 | var interpolatedValues = resampledIndices.Select(interpolation.Interpolate);
|
---|
| 114 |
|
---|
| 115 | return DoubleVector.Build.DenseOfEnumerable(interpolatedValues);
|
---|
| 116 | }
|
---|
| 117 | public static (DoubleVector, DoubleVector) ResampleToLonger(DoubleVector lhs, DoubleVector rhs) {
|
---|
| 118 | var maxLength = Math.Max(lhs.Count, rhs.Count);
|
---|
| 119 | return (ResampleToLength(lhs, maxLength), ResampleToLength(rhs, maxLength));
|
---|
| 120 | }
|
---|
| 121 | public static (DoubleVector, DoubleVector) ResampleToShorter(DoubleVector lhs, DoubleVector rhs) {
|
---|
| 122 | var minLength = Math.Min(lhs.Count, rhs.Count);
|
---|
| 123 | return (ResampleToLength(lhs, minLength), ResampleToLength(rhs, minLength));
|
---|
| 124 | }
|
---|
| 125 |
|
---|
| 126 | public static (DoubleVector, DoubleVector) CycleShorter(DoubleVector lhs, DoubleVector rhs) {
|
---|
| 127 | var targetLength = Math.Max(lhs.Count, rhs.Count);
|
---|
| 128 |
|
---|
| 129 | DoubleVector CycleVector(DoubleVector v) {
|
---|
| 130 | if (v.Count == targetLength) return v;
|
---|
| 131 | var cycledValues = Enumerable.Range(0, targetLength).Select(i => v[i % v.Count]);
|
---|
| 132 | return DoubleVector.Build.DenseOfEnumerable(cycledValues);
|
---|
| 133 | }
|
---|
| 134 |
|
---|
| 135 | return (CycleVector(lhs), CycleVector(rhs));
|
---|
| 136 | }
|
---|
| 137 | #endregion
|
---|
| 138 |
|
---|
| 139 | public static (DoubleVector lhs, DoubleVector rhs) ApplyVectorLengthStrategy(VectorLengthStrategy strategy, DoubleVector lhs, DoubleVector rhs,
|
---|
| 140 | double neutralElement = double.NaN) {
|
---|
| 141 |
|
---|
| 142 | switch (strategy) {
|
---|
| 143 | case VectorLengthStrategy.ExceptionIfDifferent: return ExceptionIfDifferent(lhs, rhs);
|
---|
| 144 | case VectorLengthStrategy.FillShorterWithNaN: return FillShorter(lhs, rhs, double.NaN);
|
---|
| 145 | case VectorLengthStrategy.FillShorterWithNeutralElement: return FillShorter(lhs, rhs, neutralElement);
|
---|
| 146 | case VectorLengthStrategy.CutLonger: return CutLonger(lhs, rhs);
|
---|
| 147 | case VectorLengthStrategy.ResampleToLonger: return ResampleToLonger(lhs, rhs);
|
---|
| 148 | case VectorLengthStrategy.ResampleToShorter: return ResampleToShorter(lhs, rhs);
|
---|
| 149 | case VectorLengthStrategy.CycleShorter: return CycleShorter(lhs, rhs);
|
---|
| 150 | default: throw new ArgumentOutOfRangeException(nameof(strategy), strategy, null);
|
---|
| 151 | }
|
---|
| 152 | }
|
---|
| 153 |
|
---|
[17604] | 154 | #region Aggregation Symbols
|
---|
| 155 | private static Type[] AggregationSymbols = new[] {
|
---|
| 156 | typeof(Sum), typeof(Mean), typeof(Length), typeof(StandardDeviation), typeof(Variance),
|
---|
| 157 | typeof(EuclideanDistance), typeof(Covariance)
|
---|
| 158 | };
|
---|
| 159 | #endregion
|
---|
| 160 |
|
---|
[7615] | 161 | private const string EvaluatedSolutionsParameterName = "EvaluatedSolutions";
|
---|
[17467] | 162 | private const string FinalAggregationParameterName = "FinalAggregation";
|
---|
[17721] | 163 | private const string DifferentVectorLengthStrategyParameterName = "DifferentVectorLengthStrategy";
|
---|
[5571] | 164 |
|
---|
[13248] | 165 | public override bool CanChangeName {
|
---|
| 166 | get { return false; }
|
---|
| 167 | }
|
---|
[5571] | 168 |
|
---|
[13248] | 169 | public override bool CanChangeDescription {
|
---|
| 170 | get { return false; }
|
---|
| 171 | }
|
---|
| 172 |
|
---|
[5749] | 173 | #region parameter properties
|
---|
[13248] | 174 | public IFixedValueParameter<IntValue> EvaluatedSolutionsParameter {
|
---|
| 175 | get { return (IFixedValueParameter<IntValue>)Parameters[EvaluatedSolutionsParameterName]; }
|
---|
[7615] | 176 | }
|
---|
[17467] | 177 | public IFixedValueParameter<EnumValue<Aggregation>> FinalAggregationParameter {
|
---|
| 178 | get { return (IFixedValueParameter<EnumValue<Aggregation>>)Parameters[FinalAggregationParameterName]; }
|
---|
| 179 | }
|
---|
[17721] | 180 | public IFixedValueParameter<EnumValue<VectorLengthStrategy>> DifferentVectorLengthStrategyParameter {
|
---|
| 181 | get { return (IFixedValueParameter<EnumValue<VectorLengthStrategy>>)Parameters[DifferentVectorLengthStrategyParameterName]; }
|
---|
| 182 | }
|
---|
[5749] | 183 | #endregion
|
---|
| 184 |
|
---|
| 185 | #region properties
|
---|
[13248] | 186 | public int EvaluatedSolutions {
|
---|
| 187 | get { return EvaluatedSolutionsParameter.Value.Value; }
|
---|
| 188 | set { EvaluatedSolutionsParameter.Value.Value = value; }
|
---|
[7615] | 189 | }
|
---|
[17467] | 190 | public Aggregation FinalAggregation {
|
---|
| 191 | get { return FinalAggregationParameter.Value.Value; }
|
---|
| 192 | set { FinalAggregationParameter.Value.Value = value; }
|
---|
| 193 | }
|
---|
[17721] | 194 | public VectorLengthStrategy DifferentVectorLengthStrategy {
|
---|
| 195 | get { return DifferentVectorLengthStrategyParameter.Value.Value; }
|
---|
| 196 | set { DifferentVectorLengthStrategyParameter.Value.Value = value; }
|
---|
| 197 | }
|
---|
[5749] | 198 | #endregion
|
---|
| 199 |
|
---|
[5571] | 200 | [StorableConstructor]
|
---|
[17455] | 201 | protected SymbolicDataAnalysisExpressionTreeVectorInterpreter(StorableConstructorFlag _) : base(_) { }
|
---|
[13248] | 202 |
|
---|
[17455] | 203 | protected SymbolicDataAnalysisExpressionTreeVectorInterpreter(SymbolicDataAnalysisExpressionTreeVectorInterpreter original, Cloner cloner)
|
---|
[13251] | 204 | : base(original, cloner) { }
|
---|
[13248] | 205 |
|
---|
[5571] | 206 | public override IDeepCloneable Clone(Cloner cloner) {
|
---|
[17455] | 207 | return new SymbolicDataAnalysisExpressionTreeVectorInterpreter(this, cloner);
|
---|
[5571] | 208 | }
|
---|
| 209 |
|
---|
[17455] | 210 | public SymbolicDataAnalysisExpressionTreeVectorInterpreter()
|
---|
[17830] | 211 | : this("SymbolicDataAnalysisExpressionTreeVectorInterpreter", "Interpreter for symbolic expression trees including vector arithmetic.") { }
|
---|
[5571] | 212 |
|
---|
[17455] | 213 | protected SymbolicDataAnalysisExpressionTreeVectorInterpreter(string name, string description)
|
---|
[8436] | 214 | : base(name, description) {
|
---|
[13248] | 215 | Parameters.Add(new FixedValueParameter<IntValue>(EvaluatedSolutionsParameterName, "A counter for the total number of solutions the interpreter has evaluated", new IntValue(0)));
|
---|
[17467] | 216 | Parameters.Add(new FixedValueParameter<EnumValue<Aggregation>>(FinalAggregationParameterName, "If root node of the expression tree results in a Vector it is aggregated according to this parameter", new EnumValue<Aggregation>(Aggregation.Mean)));
|
---|
[17721] | 217 | Parameters.Add(new FixedValueParameter<EnumValue<VectorLengthStrategy>>(DifferentVectorLengthStrategyParameterName, "", new EnumValue<VectorLengthStrategy>(VectorLengthStrategy.ExceptionIfDifferent)));
|
---|
[8436] | 218 | }
|
---|
| 219 |
|
---|
[7615] | 220 | [StorableHook(HookType.AfterDeserialization)]
|
---|
| 221 | private void AfterDeserialization() {
|
---|
[17467] | 222 | if (!Parameters.ContainsKey(FinalAggregationParameterName)) {
|
---|
| 223 | Parameters.Add(new FixedValueParameter<EnumValue<Aggregation>>(FinalAggregationParameterName, "If root node of the expression tree results in a Vector it is aggregated according to this parameter", new EnumValue<Aggregation>(Aggregation.Mean)));
|
---|
| 224 | }
|
---|
[17721] | 225 | if (!Parameters.ContainsKey(DifferentVectorLengthStrategyParameterName)) {
|
---|
| 226 | Parameters.Add(new FixedValueParameter<EnumValue<VectorLengthStrategy>>(DifferentVectorLengthStrategyParameterName, "", new EnumValue<VectorLengthStrategy>(VectorLengthStrategy.ExceptionIfDifferent)));
|
---|
| 227 | }
|
---|
[7615] | 228 | }
|
---|
| 229 |
|
---|
| 230 | #region IStatefulItem
|
---|
| 231 | public void InitializeState() {
|
---|
[13248] | 232 | EvaluatedSolutions = 0;
|
---|
[7615] | 233 | }
|
---|
| 234 |
|
---|
[13248] | 235 | public void ClearState() { }
|
---|
[7615] | 236 | #endregion
|
---|
| 237 |
|
---|
[13251] | 238 | private readonly object syncRoot = new object();
|
---|
[17455] | 239 | public IEnumerable<double> GetSymbolicExpressionTreeValues(ISymbolicExpressionTree tree, IDataset dataset, IEnumerable<int> rows) {
|
---|
[13251] | 240 | lock (syncRoot) {
|
---|
[13248] | 241 | EvaluatedSolutions++; // increment the evaluated solutions counter
|
---|
[9004] | 242 | }
|
---|
[8436] | 243 | var state = PrepareInterpreterState(tree, dataset);
|
---|
| 244 |
|
---|
| 245 | foreach (var rowEnum in rows) {
|
---|
| 246 | int row = rowEnum;
|
---|
[17455] | 247 | var result = Evaluate(dataset, ref row, state);
|
---|
[17463] | 248 | if (result.IsScalar)
|
---|
| 249 | yield return result.Scalar;
|
---|
[17467] | 250 | else if (result.IsVector) {
|
---|
[17721] | 251 | yield return Aggregate(FinalAggregation, result.Vector);
|
---|
[17467] | 252 | } else
|
---|
[17463] | 253 | yield return double.NaN;
|
---|
[8436] | 254 | state.Reset();
|
---|
| 255 | }
|
---|
[7154] | 256 | }
|
---|
| 257 |
|
---|
[17726] | 258 | public IEnumerable<Dictionary<ISymbolicExpressionTreeNode, EvaluationResult>> GetIntermediateNodeValues(ISymbolicExpressionTree tree, IDataset dataset, IEnumerable<int> rows) {
|
---|
| 259 | var state = PrepareInterpreterState(tree, dataset);
|
---|
| 260 |
|
---|
| 261 | foreach (var rowEnum in rows) {
|
---|
| 262 | int row = rowEnum;
|
---|
| 263 | var traceDict = new Dictionary<ISymbolicExpressionTreeNode, EvaluationResult>();
|
---|
| 264 | var result = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 265 | traceDict.Add(tree.Root.GetSubtree(0), result); // Add StartSymbol
|
---|
| 266 | yield return traceDict;
|
---|
| 267 | state.Reset();
|
---|
| 268 | }
|
---|
| 269 | }
|
---|
| 270 |
|
---|
[12509] | 271 | private static InterpreterState PrepareInterpreterState(ISymbolicExpressionTree tree, IDataset dataset) {
|
---|
[8436] | 272 | Instruction[] code = SymbolicExpressionTreeCompiler.Compile(tree, OpCodes.MapSymbolToOpCode);
|
---|
[5987] | 273 | int necessaryArgStackSize = 0;
|
---|
[8436] | 274 | foreach (Instruction instr in code) {
|
---|
[6860] | 275 | if (instr.opCode == OpCodes.Variable) {
|
---|
[8436] | 276 | var variableTreeNode = (VariableTreeNode)instr.dynamicNode;
|
---|
[17460] | 277 | if (dataset.VariableHasType<double>(variableTreeNode.VariableName))
|
---|
| 278 | instr.data = dataset.GetReadOnlyDoubleValues(variableTreeNode.VariableName);
|
---|
| 279 | else if (dataset.VariableHasType<DoubleVector>(variableTreeNode.VariableName))
|
---|
| 280 | instr.data = dataset.GetReadOnlyDoubleVectorValues(variableTreeNode.VariableName);
|
---|
| 281 | else throw new NotSupportedException($"Type of variable {variableTreeNode.VariableName} is not supported.");
|
---|
[14826] | 282 | } else if (instr.opCode == OpCodes.FactorVariable) {
|
---|
| 283 | var factorTreeNode = instr.dynamicNode as FactorVariableTreeNode;
|
---|
| 284 | instr.data = dataset.GetReadOnlyStringValues(factorTreeNode.VariableName);
|
---|
| 285 | } else if (instr.opCode == OpCodes.BinaryFactorVariable) {
|
---|
| 286 | var factorTreeNode = instr.dynamicNode as BinaryFactorVariableTreeNode;
|
---|
| 287 | instr.data = dataset.GetReadOnlyStringValues(factorTreeNode.VariableName);
|
---|
[5571] | 288 | } else if (instr.opCode == OpCodes.LagVariable) {
|
---|
[8436] | 289 | var laggedVariableTreeNode = (LaggedVariableTreeNode)instr.dynamicNode;
|
---|
[9828] | 290 | instr.data = dataset.GetReadOnlyDoubleValues(laggedVariableTreeNode.VariableName);
|
---|
[6860] | 291 | } else if (instr.opCode == OpCodes.VariableCondition) {
|
---|
[8436] | 292 | var variableConditionTreeNode = (VariableConditionTreeNode)instr.dynamicNode;
|
---|
[9828] | 293 | instr.data = dataset.GetReadOnlyDoubleValues(variableConditionTreeNode.VariableName);
|
---|
[5987] | 294 | } else if (instr.opCode == OpCodes.Call) {
|
---|
| 295 | necessaryArgStackSize += instr.nArguments + 1;
|
---|
[5571] | 296 | }
|
---|
| 297 | }
|
---|
[8436] | 298 | return new InterpreterState(code, necessaryArgStackSize);
|
---|
| 299 | }
|
---|
[5571] | 300 |
|
---|
[17455] | 301 |
|
---|
| 302 | public struct EvaluationResult {
|
---|
| 303 | public double Scalar { get; }
|
---|
| 304 | public bool IsScalar => !double.IsNaN(Scalar);
|
---|
| 305 |
|
---|
| 306 | public DoubleVector Vector { get; }
|
---|
[17465] | 307 | public bool IsVector => !(Vector.Count == 1 && double.IsNaN(Vector[0]));
|
---|
[17455] | 308 |
|
---|
| 309 | public bool IsNaN => !IsScalar && !IsVector;
|
---|
| 310 |
|
---|
| 311 | public EvaluationResult(double scalar) {
|
---|
| 312 | Scalar = scalar;
|
---|
[17465] | 313 | Vector = NaNVector;
|
---|
[17455] | 314 | }
|
---|
| 315 | public EvaluationResult(DoubleVector vector) {
|
---|
[17463] | 316 | if (vector == null) throw new ArgumentNullException(nameof(vector));
|
---|
[17455] | 317 | Vector = vector;
|
---|
| 318 | Scalar = double.NaN;
|
---|
| 319 | }
|
---|
| 320 |
|
---|
| 321 | public override string ToString() {
|
---|
| 322 | if (IsScalar) return Scalar.ToString();
|
---|
| 323 | if (IsVector) return Vector.ToVectorString();
|
---|
| 324 | return "NaN";
|
---|
| 325 | }
|
---|
| 326 |
|
---|
[17465] | 327 | private static readonly DoubleVector NaNVector = DoubleVector.Build.Dense(1, double.NaN);
|
---|
[17455] | 328 | public static readonly EvaluationResult NaN = new EvaluationResult(double.NaN);
|
---|
| 329 | }
|
---|
| 330 |
|
---|
| 331 | private static EvaluationResult ArithmeticApply(EvaluationResult lhs, EvaluationResult rhs,
|
---|
[17721] | 332 | Func<DoubleVector, DoubleVector, (DoubleVector, DoubleVector)> lengthStrategy,
|
---|
[17455] | 333 | Func<double, double, double> ssFunc = null,
|
---|
| 334 | Func<double, DoubleVector, DoubleVector> svFunc = null,
|
---|
| 335 | Func<DoubleVector, double, DoubleVector> vsFunc = null,
|
---|
| 336 | Func<DoubleVector, DoubleVector, DoubleVector> vvFunc = null) {
|
---|
[17721] | 337 |
|
---|
[17455] | 338 | if (lhs.IsScalar && rhs.IsScalar && ssFunc != null) return new EvaluationResult(ssFunc(lhs.Scalar, rhs.Scalar));
|
---|
| 339 | if (lhs.IsScalar && rhs.IsVector && svFunc != null) return new EvaluationResult(svFunc(lhs.Scalar, rhs.Vector));
|
---|
| 340 | if (lhs.IsVector && rhs.IsScalar && vsFunc != null) return new EvaluationResult(vsFunc(lhs.Vector, rhs.Scalar));
|
---|
[17721] | 341 | if (lhs.IsVector && rhs.IsVector && vvFunc != null) {
|
---|
| 342 | if (lhs.Vector.Count == rhs.Vector.Count) {
|
---|
| 343 | return new EvaluationResult(vvFunc(lhs.Vector, rhs.Vector));
|
---|
| 344 | } else {
|
---|
| 345 | var (lhsVector, rhsVector) = lengthStrategy(lhs.Vector, rhs.Vector);
|
---|
| 346 | return new EvaluationResult(vvFunc(lhsVector, rhsVector));
|
---|
| 347 | }
|
---|
| 348 | }
|
---|
[17463] | 349 | return EvaluationResult.NaN;
|
---|
[17455] | 350 | }
|
---|
| 351 |
|
---|
| 352 | private static EvaluationResult FunctionApply(EvaluationResult val,
|
---|
| 353 | Func<double, double> sFunc = null,
|
---|
| 354 | Func<DoubleVector, DoubleVector> vFunc = null) {
|
---|
| 355 | if (val.IsScalar && sFunc != null) return new EvaluationResult(sFunc(val.Scalar));
|
---|
| 356 | if (val.IsVector && vFunc != null) return new EvaluationResult(vFunc(val.Vector));
|
---|
[17463] | 357 | return EvaluationResult.NaN;
|
---|
[17455] | 358 | }
|
---|
[17460] | 359 | private static EvaluationResult AggregateApply(EvaluationResult val,
|
---|
| 360 | Func<double, double> sFunc = null,
|
---|
| 361 | Func<DoubleVector, double> vFunc = null) {
|
---|
| 362 | if (val.IsScalar && sFunc != null) return new EvaluationResult(sFunc(val.Scalar));
|
---|
| 363 | if (val.IsVector && vFunc != null) return new EvaluationResult(vFunc(val.Vector));
|
---|
[17463] | 364 | return EvaluationResult.NaN;
|
---|
[17460] | 365 | }
|
---|
[17573] | 366 |
|
---|
[17726] | 367 | private static EvaluationResult WindowedAggregateApply(EvaluationResult val, WindowedSymbolTreeNode node,
|
---|
[17573] | 368 | Func<double, double> sFunc = null,
|
---|
| 369 | Func<DoubleVector, double> vFunc = null) {
|
---|
| 370 |
|
---|
[18058] | 371 | // Parameters are interpreted as start and end with wrapping
|
---|
| 372 | var start = node.Offset;
|
---|
| 373 | var end = node.Length;
|
---|
[17573] | 374 |
|
---|
| 375 | DoubleVector SubVector(DoubleVector v) {
|
---|
[18058] | 376 | int startIdx = (int)Math.Round(start * v.Count);
|
---|
| 377 | int endIdx = (int)Math.Round(end * v.Count);
|
---|
| 378 | int size = v.Count;
|
---|
| 379 | if (startIdx < endIdx) {
|
---|
| 380 | return v.SubVector(startIdx, count: endIdx - startIdx);
|
---|
| 381 | } else { // wrap around
|
---|
| 382 | var resultVector = DoubleVector.Build.Dense(size: size - (startIdx - endIdx));
|
---|
| 383 | v.CopySubVectorTo(resultVector, startIdx, 0, size - startIdx); // copy [startIdx:size] to [0:size-startIdx]
|
---|
| 384 | v.CopySubVectorTo(resultVector, 0, size - startIdx, endIdx); // copy [0:endIdx] to [size-startIdx:size]
|
---|
| 385 | return resultVector;
|
---|
| 386 | }
|
---|
| 387 | }
|
---|
[17573] | 388 |
|
---|
| 389 | if (val.IsScalar && sFunc != null) return new EvaluationResult(sFunc(val.Scalar));
|
---|
| 390 | if (val.IsVector && vFunc != null) return new EvaluationResult(vFunc(SubVector(val.Vector)));
|
---|
| 391 | return EvaluationResult.NaN;
|
---|
| 392 | }
|
---|
[17726] | 393 | private static EvaluationResult WindowedFunctionApply(EvaluationResult val, IWindowedSymbolTreeNode node,
|
---|
| 394 | Func<double, double> sFunc = null,
|
---|
| 395 | Func<DoubleVector, DoubleVector> vFunc = null) {
|
---|
[18058] | 396 | // Parameters are interpreted as start and end with wrapping
|
---|
| 397 | var start = node.Offset;
|
---|
| 398 | var end = node.Length;
|
---|
[17726] | 399 |
|
---|
| 400 | DoubleVector SubVector(DoubleVector v) {
|
---|
[18058] | 401 | int startIdx = (int)Math.Round(start * v.Count);
|
---|
| 402 | int endIdx = (int)Math.Round(end * v.Count);
|
---|
| 403 | int size = v.Count;
|
---|
| 404 | if (startIdx < endIdx) {
|
---|
| 405 | return v.SubVector(startIdx, count: endIdx - startIdx);
|
---|
| 406 | } else { // wrap around
|
---|
| 407 | var resultVector = DoubleVector.Build.Dense(size: size - (startIdx - endIdx));
|
---|
| 408 | v.CopySubVectorTo(resultVector, startIdx, 0, size - startIdx); // copy [startIdx:size] to [0:size-startIdx]
|
---|
| 409 | v.CopySubVectorTo(resultVector, 0, size - startIdx, endIdx); // copy [0:endIdx] to [size-startIdx:size]
|
---|
| 410 | return resultVector;
|
---|
| 411 | }
|
---|
| 412 | }
|
---|
[17726] | 413 |
|
---|
| 414 | if (val.IsScalar && sFunc != null) return new EvaluationResult(sFunc(val.Scalar));
|
---|
| 415 | if (val.IsVector && vFunc != null) return new EvaluationResult(vFunc(SubVector(val.Vector)));
|
---|
| 416 | return EvaluationResult.NaN;
|
---|
| 417 | }
|
---|
| 418 |
|
---|
[17554] | 419 | private static EvaluationResult AggregateMultipleApply(EvaluationResult lhs, EvaluationResult rhs,
|
---|
[17721] | 420 | Func<DoubleVector, DoubleVector, (DoubleVector, DoubleVector)> lengthStrategy,
|
---|
[17554] | 421 | Func<double, double, double> ssFunc = null,
|
---|
| 422 | Func<double, DoubleVector, double> svFunc = null,
|
---|
| 423 | Func<DoubleVector, double, double> vsFunc = null,
|
---|
| 424 | Func<DoubleVector, DoubleVector, double> vvFunc = null) {
|
---|
| 425 | if (lhs.IsScalar && rhs.IsScalar && ssFunc != null) return new EvaluationResult(ssFunc(lhs.Scalar, rhs.Scalar));
|
---|
| 426 | if (lhs.IsScalar && rhs.IsVector && svFunc != null) return new EvaluationResult(svFunc(lhs.Scalar, rhs.Vector));
|
---|
| 427 | if (lhs.IsVector && rhs.IsScalar && vsFunc != null) return new EvaluationResult(vsFunc(lhs.Vector, rhs.Scalar));
|
---|
[17721] | 428 | if (lhs.IsVector && rhs.IsVector && vvFunc != null) {
|
---|
| 429 | if (lhs.Vector.Count == rhs.Vector.Count) {
|
---|
| 430 | return new EvaluationResult(vvFunc(lhs.Vector, rhs.Vector));
|
---|
| 431 | } else {
|
---|
| 432 | var (lhsVector, rhsVector) = lengthStrategy(lhs.Vector, rhs.Vector);
|
---|
| 433 | return new EvaluationResult(vvFunc(lhsVector, rhsVector));
|
---|
| 434 | }
|
---|
| 435 | }
|
---|
[17554] | 436 | return EvaluationResult.NaN;
|
---|
| 437 | }
|
---|
[17455] | 438 |
|
---|
[17604] | 439 | public virtual Type GetNodeType(ISymbolicExpressionTreeNode node) {
|
---|
| 440 | if (node.DataType != null)
|
---|
| 441 | return node.DataType;
|
---|
[17593] | 442 |
|
---|
[17604] | 443 | if (AggregationSymbols.Contains(node.Symbol.GetType()))
|
---|
| 444 | return typeof(double);
|
---|
[17593] | 445 |
|
---|
[17604] | 446 | var argumentTypes = node.Subtrees.Select(GetNodeType);
|
---|
| 447 | if (argumentTypes.Any(t => t == typeof(DoubleVector)))
|
---|
| 448 | return typeof(DoubleVector);
|
---|
[17593] | 449 |
|
---|
[17604] | 450 | return typeof(double);
|
---|
[17593] | 451 | }
|
---|
| 452 |
|
---|
[17726] | 453 |
|
---|
| 454 | public virtual EvaluationResult Evaluate(IDataset dataset, ref int row, InterpreterState state,
|
---|
| 455 | IDictionary<ISymbolicExpressionTreeNode, EvaluationResult> traceDict = null) {
|
---|
| 456 |
|
---|
| 457 | void TraceEvaluation(Instruction instr, EvaluationResult result) {
|
---|
| 458 | traceDict?.Add(instr.dynamicNode, result);
|
---|
| 459 | }
|
---|
| 460 |
|
---|
[5571] | 461 | Instruction currentInstr = state.NextInstruction();
|
---|
| 462 | switch (currentInstr.opCode) {
|
---|
| 463 | case OpCodes.Add: {
|
---|
[17726] | 464 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
[5571] | 465 | for (int i = 1; i < currentInstr.nArguments; i++) {
|
---|
[17726] | 466 | var op = Evaluate(dataset, ref row, state, traceDict);
|
---|
[17455] | 467 | cur = ArithmeticApply(cur, op,
|
---|
[17721] | 468 | (lhs, rhs) => ApplyVectorLengthStrategy(DifferentVectorLengthStrategy, lhs, rhs, 0.0),
|
---|
[17455] | 469 | (s1, s2) => s1 + s2,
|
---|
| 470 | (s1, v2) => s1 + v2,
|
---|
| 471 | (v1, s2) => v1 + s2,
|
---|
| 472 | (v1, v2) => v1 + v2);
|
---|
[5571] | 473 | }
|
---|
[17726] | 474 | TraceEvaluation(currentInstr, cur);
|
---|
[17455] | 475 | return cur;
|
---|
[5571] | 476 | }
|
---|
| 477 | case OpCodes.Sub: {
|
---|
[17726] | 478 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
[5571] | 479 | for (int i = 1; i < currentInstr.nArguments; i++) {
|
---|
[17726] | 480 | var op = Evaluate(dataset, ref row, state, traceDict);
|
---|
[17455] | 481 | cur = ArithmeticApply(cur, op,
|
---|
[17721] | 482 | (lhs, rhs) => ApplyVectorLengthStrategy(DifferentVectorLengthStrategy, lhs, rhs, 0.0),
|
---|
[17455] | 483 | (s1, s2) => s1 - s2,
|
---|
| 484 | (s1, v2) => s1 - v2,
|
---|
| 485 | (v1, s2) => v1 - s2,
|
---|
| 486 | (v1, v2) => v1 - v2);
|
---|
[5571] | 487 | }
|
---|
[17786] | 488 | if (currentInstr.nArguments == 1)
|
---|
| 489 | cur = FunctionApply(cur,
|
---|
| 490 | s => -s,
|
---|
| 491 | v => -v);
|
---|
[17726] | 492 | TraceEvaluation(currentInstr, cur);
|
---|
[17455] | 493 | return cur;
|
---|
[5571] | 494 | }
|
---|
| 495 | case OpCodes.Mul: {
|
---|
[17726] | 496 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
[5571] | 497 | for (int i = 1; i < currentInstr.nArguments; i++) {
|
---|
[17726] | 498 | var op = Evaluate(dataset, ref row, state, traceDict);
|
---|
[17455] | 499 | cur = ArithmeticApply(cur, op,
|
---|
[17721] | 500 | (lhs, rhs) => ApplyVectorLengthStrategy(DifferentVectorLengthStrategy, lhs, rhs, 1.0),
|
---|
[17455] | 501 | (s1, s2) => s1 * s2,
|
---|
| 502 | (s1, v2) => s1 * v2,
|
---|
| 503 | (v1, s2) => v1 * s2,
|
---|
| 504 | (v1, v2) => v1.PointwiseMultiply(v2));
|
---|
[5571] | 505 | }
|
---|
[17726] | 506 | TraceEvaluation(currentInstr, cur);
|
---|
[17455] | 507 | return cur;
|
---|
[5571] | 508 | }
|
---|
| 509 | case OpCodes.Div: {
|
---|
[17726] | 510 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
[5571] | 511 | for (int i = 1; i < currentInstr.nArguments; i++) {
|
---|
[17726] | 512 | var op = Evaluate(dataset, ref row, state, traceDict);
|
---|
[17455] | 513 | cur = ArithmeticApply(cur, op,
|
---|
[17721] | 514 | (lhs, rhs) => ApplyVectorLengthStrategy(DifferentVectorLengthStrategy, lhs, rhs, 1.0),
|
---|
[17455] | 515 | (s1, s2) => s1 / s2,
|
---|
| 516 | (s1, v2) => s1 / v2,
|
---|
| 517 | (v1, s2) => v1 / s2,
|
---|
| 518 | (v1, v2) => v1 / v2);
|
---|
[5571] | 519 | }
|
---|
[17786] | 520 | if (currentInstr.nArguments == 1)
|
---|
| 521 | cur = FunctionApply(cur,
|
---|
| 522 | s => 1 / s,
|
---|
| 523 | v => 1 / v);
|
---|
[17726] | 524 | TraceEvaluation(currentInstr, cur);
|
---|
[17455] | 525 | return cur;
|
---|
[5571] | 526 | }
|
---|
[16356] | 527 | case OpCodes.Absolute: {
|
---|
[17726] | 528 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 529 | cur = FunctionApply(cur, Math.Abs, DoubleVector.Abs);
|
---|
| 530 | TraceEvaluation(currentInstr, cur);
|
---|
| 531 | return cur;
|
---|
[16356] | 532 | }
|
---|
[16656] | 533 | case OpCodes.Tanh: {
|
---|
[17726] | 534 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 535 | cur = FunctionApply(cur, Math.Tanh, DoubleVector.Tanh);
|
---|
| 536 | TraceEvaluation(currentInstr, cur);
|
---|
| 537 | return cur;
|
---|
[16656] | 538 | }
|
---|
[5571] | 539 | case OpCodes.Cos: {
|
---|
[17726] | 540 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 541 | cur = FunctionApply(cur, Math.Cos, DoubleVector.Cos);
|
---|
| 542 | TraceEvaluation(currentInstr, cur);
|
---|
| 543 | return cur;
|
---|
[5571] | 544 | }
|
---|
| 545 | case OpCodes.Sin: {
|
---|
[17726] | 546 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 547 | cur = FunctionApply(cur, Math.Sin, DoubleVector.Sin);
|
---|
| 548 | TraceEvaluation(currentInstr, cur);
|
---|
| 549 | return cur;
|
---|
[5571] | 550 | }
|
---|
| 551 | case OpCodes.Tan: {
|
---|
[17726] | 552 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 553 | cur = FunctionApply(cur, Math.Tan, DoubleVector.Tan);
|
---|
| 554 | TraceEvaluation(currentInstr, cur);
|
---|
| 555 | return cur;
|
---|
[5571] | 556 | }
|
---|
[7842] | 557 | case OpCodes.Square: {
|
---|
[17726] | 558 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 559 | cur = FunctionApply(cur,
|
---|
[17455] | 560 | s => Math.Pow(s, 2),
|
---|
| 561 | v => v.PointwisePower(2));
|
---|
[17726] | 562 | TraceEvaluation(currentInstr, cur);
|
---|
| 563 | return cur;
|
---|
[7842] | 564 | }
|
---|
[16356] | 565 | case OpCodes.Cube: {
|
---|
[17726] | 566 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 567 | cur = FunctionApply(cur,
|
---|
[17455] | 568 | s => Math.Pow(s, 3),
|
---|
| 569 | v => v.PointwisePower(3));
|
---|
[17726] | 570 | TraceEvaluation(currentInstr, cur);
|
---|
| 571 | return cur;
|
---|
[16356] | 572 | }
|
---|
[5571] | 573 | case OpCodes.Power: {
|
---|
[17726] | 574 | var x = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 575 | var y = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 576 | var cur = ArithmeticApply(x, y,
|
---|
[17721] | 577 | (lhs, rhs) => lhs.Count < rhs.Count
|
---|
| 578 | ? CutLonger(lhs, rhs)
|
---|
| 579 | : ApplyVectorLengthStrategy(DifferentVectorLengthStrategy, lhs, rhs, 1.0),
|
---|
[17455] | 580 | (s1, s2) => Math.Pow(s1, Math.Round(s2)),
|
---|
| 581 | (s1, v2) => DoubleVector.Build.Dense(v2.Count, s1).PointwisePower(DoubleVector.Round(v2)),
|
---|
| 582 | (v1, s2) => v1.PointwisePower(Math.Round(s2)),
|
---|
| 583 | (v1, v2) => v1.PointwisePower(DoubleVector.Round(v2)));
|
---|
[17726] | 584 | TraceEvaluation(currentInstr, cur);
|
---|
| 585 | return cur;
|
---|
[5571] | 586 | }
|
---|
[7842] | 587 | case OpCodes.SquareRoot: {
|
---|
[17726] | 588 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 589 | cur = FunctionApply(cur,
|
---|
[17455] | 590 | s => Math.Sqrt(s),
|
---|
| 591 | v => DoubleVector.Sqrt(v));
|
---|
[17726] | 592 | TraceEvaluation(currentInstr, cur);
|
---|
| 593 | return cur;
|
---|
[7842] | 594 | }
|
---|
[16356] | 595 | case OpCodes.CubeRoot: {
|
---|
[17726] | 596 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 597 | cur = FunctionApply(cur,
|
---|
[17455] | 598 | s => s < 0 ? -Math.Pow(-s, 1.0 / 3.0) : Math.Pow(s, 1.0 / 3.0),
|
---|
| 599 | v => v.Map(s => s < 0 ? -Math.Pow(-s, 1.0 / 3.0) : Math.Pow(s, 1.0 / 3.0)));
|
---|
[17726] | 600 | TraceEvaluation(currentInstr, cur);
|
---|
| 601 | return cur;
|
---|
[16356] | 602 | }
|
---|
[5571] | 603 | case OpCodes.Root: {
|
---|
[17726] | 604 | var x = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 605 | var y = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 606 | var cur = ArithmeticApply(x, y,
|
---|
[17721] | 607 | (lhs, rhs) => lhs.Count < rhs.Count
|
---|
| 608 | ? CutLonger(lhs, rhs)
|
---|
| 609 | : ApplyVectorLengthStrategy(DifferentVectorLengthStrategy, lhs, rhs, 1.0),
|
---|
[17455] | 610 | (s1, s2) => Math.Pow(s1, 1.0 / Math.Round(s2)),
|
---|
| 611 | (s1, v2) => DoubleVector.Build.Dense(v2.Count, s1).PointwisePower(1.0 / DoubleVector.Round(v2)),
|
---|
| 612 | (v1, s2) => v1.PointwisePower(1.0 / Math.Round(s2)),
|
---|
| 613 | (v1, v2) => v1.PointwisePower(1.0 / DoubleVector.Round(v2)));
|
---|
[17726] | 614 | TraceEvaluation(currentInstr, cur);
|
---|
| 615 | return cur;
|
---|
[5571] | 616 | }
|
---|
| 617 | case OpCodes.Exp: {
|
---|
[17726] | 618 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 619 | cur = FunctionApply(cur,
|
---|
[17455] | 620 | s => Math.Exp(s),
|
---|
| 621 | v => DoubleVector.Exp(v));
|
---|
[17726] | 622 | TraceEvaluation(currentInstr, cur);
|
---|
| 623 | return cur;
|
---|
[5571] | 624 | }
|
---|
| 625 | case OpCodes.Log: {
|
---|
[17726] | 626 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 627 | cur = FunctionApply(cur,
|
---|
[17455] | 628 | s => Math.Log(s),
|
---|
| 629 | v => DoubleVector.Log(v));
|
---|
[17726] | 630 | TraceEvaluation(currentInstr, cur);
|
---|
| 631 | return cur;
|
---|
[5571] | 632 | }
|
---|
[17460] | 633 | case OpCodes.Sum: {
|
---|
[17726] | 634 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 635 | cur = AggregateApply(cur,
|
---|
[17460] | 636 | s => s,
|
---|
| 637 | v => v.Sum());
|
---|
[17726] | 638 | TraceEvaluation(currentInstr, cur);
|
---|
| 639 | return cur;
|
---|
[17460] | 640 | }
|
---|
[17466] | 641 | case OpCodes.Mean: {
|
---|
[17726] | 642 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 643 | cur = AggregateApply(cur,
|
---|
[17460] | 644 | s => s,
|
---|
[17721] | 645 | v => Statistics.Mean(v));
|
---|
[17726] | 646 | TraceEvaluation(currentInstr, cur);
|
---|
| 647 | return cur;
|
---|
[17460] | 648 | }
|
---|
[17463] | 649 | case OpCodes.StandardDeviation: {
|
---|
[17726] | 650 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 651 | cur = AggregateApply(cur,
|
---|
[17602] | 652 | s => 0,
|
---|
| 653 | v => Statistics.PopulationStandardDeviation(v));
|
---|
[17726] | 654 | TraceEvaluation(currentInstr, cur);
|
---|
| 655 | return cur;
|
---|
[17463] | 656 | }
|
---|
[17554] | 657 | case OpCodes.Length: {
|
---|
[17726] | 658 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 659 | cur = AggregateApply(cur,
|
---|
[17593] | 660 | s => 1,
|
---|
| 661 | v => v.Count);
|
---|
[17726] | 662 | TraceEvaluation(currentInstr, cur);
|
---|
| 663 | return cur;
|
---|
[17593] | 664 | }
|
---|
[17554] | 665 | case OpCodes.Min: {
|
---|
[17726] | 666 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 667 | cur = AggregateApply(cur,
|
---|
[17593] | 668 | s => s,
|
---|
| 669 | v => Statistics.Minimum(v));
|
---|
[17726] | 670 | TraceEvaluation(currentInstr, cur);
|
---|
| 671 | return cur;
|
---|
[17593] | 672 | }
|
---|
[17554] | 673 | case OpCodes.Max: {
|
---|
[17726] | 674 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 675 | cur = AggregateApply(cur,
|
---|
[17593] | 676 | s => s,
|
---|
| 677 | v => Statistics.Maximum(v));
|
---|
[17726] | 678 | TraceEvaluation(currentInstr, cur);
|
---|
| 679 | return cur;
|
---|
[17593] | 680 | }
|
---|
[17554] | 681 | case OpCodes.Variance: {
|
---|
[17726] | 682 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 683 | cur = AggregateApply(cur,
|
---|
[17602] | 684 | s => 0,
|
---|
| 685 | v => Statistics.PopulationVariance(v));
|
---|
[17726] | 686 | TraceEvaluation(currentInstr, cur);
|
---|
| 687 | return cur;
|
---|
[17593] | 688 | }
|
---|
[17554] | 689 | case OpCodes.Skewness: {
|
---|
[17726] | 690 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 691 | cur = AggregateApply(cur,
|
---|
[17593] | 692 | s => double.NaN,
|
---|
[17602] | 693 | v => Statistics.PopulationSkewness(v));
|
---|
[17726] | 694 | TraceEvaluation(currentInstr, cur);
|
---|
| 695 | return cur;
|
---|
[17593] | 696 | }
|
---|
[17554] | 697 | case OpCodes.Kurtosis: {
|
---|
[17726] | 698 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 699 | cur = AggregateApply(cur,
|
---|
[17593] | 700 | s => double.NaN,
|
---|
[17602] | 701 | v => Statistics.PopulationKurtosis(v));
|
---|
[17726] | 702 | TraceEvaluation(currentInstr, cur);
|
---|
| 703 | return cur;
|
---|
[17593] | 704 | }
|
---|
[17554] | 705 | case OpCodes.EuclideanDistance: {
|
---|
[17726] | 706 | var x1 = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 707 | var x2 = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 708 | var cur = AggregateMultipleApply(x1, x2,
|
---|
[17721] | 709 | (lhs, rhs) => ApplyVectorLengthStrategy(DifferentVectorLengthStrategy, lhs, rhs, 0.0),
|
---|
| 710 | (s1, s2) => s1 - s2,
|
---|
| 711 | (s1, v2) => Math.Sqrt((s1 - v2).PointwisePower(2).Sum()),
|
---|
| 712 | (v1, s2) => Math.Sqrt((v1 - s2).PointwisePower(2).Sum()),
|
---|
| 713 | (v1, v2) => Math.Sqrt((v1 - v2).PointwisePower(2).Sum()));
|
---|
[17726] | 714 | TraceEvaluation(currentInstr, cur);
|
---|
| 715 | return cur;
|
---|
[17593] | 716 | }
|
---|
[17554] | 717 | case OpCodes.Covariance: {
|
---|
[17726] | 718 | var x1 = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 719 | var x2 = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 720 | var cur = AggregateMultipleApply(x1, x2,
|
---|
[17721] | 721 | (lhs, rhs) => ApplyVectorLengthStrategy(DifferentVectorLengthStrategy, lhs, rhs, 0.0),
|
---|
| 722 | (s1, s2) => 0,
|
---|
| 723 | (s1, v2) => 0,
|
---|
| 724 | (v1, s2) => 0,
|
---|
| 725 | (v1, v2) => Statistics.PopulationCovariance(v1, v2));
|
---|
[17726] | 726 | TraceEvaluation(currentInstr, cur);
|
---|
| 727 | return cur;
|
---|
[17593] | 728 | }
|
---|
[17726] | 729 | case OpCodes.SubVector: {
|
---|
| 730 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 731 | return WindowedFunctionApply(cur, (WindowedSymbolTreeNode)currentInstr.dynamicNode,
|
---|
| 732 | s => s,
|
---|
| 733 | v => v);
|
---|
| 734 | }
|
---|
[5571] | 735 | case OpCodes.Variable: {
|
---|
[17455] | 736 | if (row < 0 || row >= dataset.Rows) return EvaluationResult.NaN;
|
---|
[6740] | 737 | var variableTreeNode = (VariableTreeNode)currentInstr.dynamicNode;
|
---|
[17726] | 738 | if (currentInstr.data is IList<double> doubleList) {
|
---|
| 739 | var cur = new EvaluationResult(doubleList[row] * variableTreeNode.Weight);
|
---|
| 740 | TraceEvaluation(currentInstr, cur);
|
---|
| 741 | return cur;
|
---|
| 742 | }
|
---|
| 743 | if (currentInstr.data is IList<DoubleVector> doubleVectorList) {
|
---|
| 744 | var cur = new EvaluationResult(doubleVectorList[row] * variableTreeNode.Weight);
|
---|
| 745 | TraceEvaluation(currentInstr, cur);
|
---|
| 746 | return cur;
|
---|
| 747 | }
|
---|
[17455] | 748 | throw new NotSupportedException($"Unsupported type of variable: {currentInstr.data.GetType().GetPrettyName()}");
|
---|
[5571] | 749 | }
|
---|
[14826] | 750 | case OpCodes.BinaryFactorVariable: {
|
---|
[17455] | 751 | if (row < 0 || row >= dataset.Rows) return EvaluationResult.NaN;
|
---|
[14826] | 752 | var factorVarTreeNode = currentInstr.dynamicNode as BinaryFactorVariableTreeNode;
|
---|
[17726] | 753 | var cur = new EvaluationResult(((IList<string>)currentInstr.data)[row] == factorVarTreeNode.VariableValue ? factorVarTreeNode.Weight : 0);
|
---|
| 754 | TraceEvaluation(currentInstr, cur);
|
---|
| 755 | return cur;
|
---|
[14826] | 756 | }
|
---|
| 757 | case OpCodes.FactorVariable: {
|
---|
[17455] | 758 | if (row < 0 || row >= dataset.Rows) return EvaluationResult.NaN;
|
---|
[14826] | 759 | var factorVarTreeNode = currentInstr.dynamicNode as FactorVariableTreeNode;
|
---|
[17726] | 760 | var cur = new EvaluationResult(factorVarTreeNode.GetValue(((IList<string>)currentInstr.data)[row]));
|
---|
| 761 | TraceEvaluation(currentInstr, cur);
|
---|
| 762 | return cur;
|
---|
[14826] | 763 | }
|
---|
[5571] | 764 | case OpCodes.Constant: {
|
---|
[8436] | 765 | var constTreeNode = (ConstantTreeNode)currentInstr.dynamicNode;
|
---|
[17726] | 766 | var cur = new EvaluationResult(constTreeNode.Value);
|
---|
| 767 | TraceEvaluation(currentInstr, cur);
|
---|
| 768 | return cur;
|
---|
[5571] | 769 | }
|
---|
| 770 |
|
---|
[17830] | 771 | #region Time Series Symbols
|
---|
| 772 | case OpCodes.Median: {
|
---|
| 773 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 774 | cur = AggregateApply(cur,
|
---|
| 775 | s => s,
|
---|
| 776 | v => Statistics.Median(v));
|
---|
| 777 | TraceEvaluation(currentInstr, cur);
|
---|
| 778 | return cur;
|
---|
| 779 | }
|
---|
| 780 | case OpCodes.Quantile: {
|
---|
| 781 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 782 | var q = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 783 | cur = AggregateApply(cur,
|
---|
| 784 | s => s,
|
---|
| 785 | v => Statistics.Quantile(v, q.Scalar));
|
---|
| 786 | TraceEvaluation(currentInstr, cur);
|
---|
| 787 | return cur;
|
---|
| 788 | }
|
---|
| 789 |
|
---|
| 790 | case OpCodes.AbsoluteEnergy: {
|
---|
| 791 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 792 | cur = AggregateApply(cur,
|
---|
| 793 | s => s * s,
|
---|
| 794 | v => v.PointwisePower(2.0).Sum());
|
---|
| 795 | TraceEvaluation(currentInstr, cur);
|
---|
| 796 | return cur;
|
---|
| 797 | }
|
---|
| 798 |
|
---|
| 799 | case OpCodes.BinnedEntropy: {
|
---|
| 800 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 801 | var m = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 802 | cur = AggregateApply(cur,
|
---|
| 803 | s => 0,
|
---|
| 804 | v => {
|
---|
[17930] | 805 | int bins = Math.Max((int)Math.Round(m.Scalar), 1);
|
---|
[17830] | 806 | double minValue = v.Minimum();
|
---|
| 807 | double maxValue = v.Maximum();
|
---|
| 808 | double intervalWidth = (maxValue - minValue) / bins;
|
---|
| 809 | int totalValues = v.Count;
|
---|
| 810 | double sum = 0;
|
---|
| 811 | for (int i = 0; i < Math.Max(bins, v.Count); i++) {
|
---|
| 812 | double binMin = minValue * i;
|
---|
| 813 | double binMax = binMin + intervalWidth;
|
---|
[17930] | 814 | double countBin = v.Map(e => (e > binMin && e < binMax) ? 1.0 : 0.0).Sum();
|
---|
[17830] | 815 | double percBin = countBin / totalValues;
|
---|
| 816 | sum += percBin * Math.Log(percBin);
|
---|
| 817 | }
|
---|
| 818 |
|
---|
| 819 | return sum;
|
---|
| 820 | });
|
---|
| 821 | TraceEvaluation(currentInstr, cur);
|
---|
| 822 | return cur;
|
---|
| 823 | }
|
---|
| 824 | case OpCodes.HasLargeStandardDeviation: {
|
---|
| 825 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 826 | cur = AggregateApply(cur,
|
---|
| 827 | s => 0,
|
---|
[17930] | 828 | v => Statistics.PopulationStandardDeviation(v) > (Statistics.Maximum(v) - Statistics.Minimum(v)) / 2 ? 1.0 : 0.0);
|
---|
[17830] | 829 | TraceEvaluation(currentInstr, cur);
|
---|
| 830 | return cur;
|
---|
| 831 | }
|
---|
| 832 | case OpCodes.HasVarianceLargerThanStd: {
|
---|
| 833 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 834 | cur = AggregateApply(cur,
|
---|
| 835 | s => 0,
|
---|
[17930] | 836 | v => Statistics.PopulationVariance(v) > Statistics.StandardDeviation(v) ? 1.0 : 0.0);
|
---|
[17830] | 837 | TraceEvaluation(currentInstr, cur);
|
---|
| 838 | return cur;
|
---|
| 839 | }
|
---|
| 840 | case OpCodes.IsSymmetricLooking: {
|
---|
| 841 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 842 | cur = AggregateApply(cur,
|
---|
| 843 | s => 0,
|
---|
[17930] | 844 | v => Math.Abs(Statistics.Mean(v) - Statistics.Median(v)) < (Statistics.Maximum(v) - Statistics.Minimum(v)) / 2 ? 1.0 : 0.0);
|
---|
[17830] | 845 | TraceEvaluation(currentInstr, cur);
|
---|
| 846 | return cur;
|
---|
| 847 | }
|
---|
| 848 | case OpCodes.NumberDataPointsAboveMean: {
|
---|
| 849 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 850 | cur = AggregateApply(cur,
|
---|
| 851 | s => 0,
|
---|
| 852 | v => {
|
---|
| 853 | double mean = Statistics.Mean(v);
|
---|
[17930] | 854 | return v.Map(e => e > mean ? 1.0 : 0.0).Sum();
|
---|
[17830] | 855 | });
|
---|
| 856 | TraceEvaluation(currentInstr, cur);
|
---|
| 857 | return cur;
|
---|
| 858 | }
|
---|
| 859 | case OpCodes.NumberDataPointsAboveMedian: {
|
---|
| 860 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 861 | cur = AggregateApply(cur,
|
---|
| 862 | s => 0,
|
---|
| 863 | v => {
|
---|
| 864 | double median = Statistics.Median(v);
|
---|
[17930] | 865 | return v.Map(e => e > median ? 1.0 : 0.0).Sum();
|
---|
[17830] | 866 | });
|
---|
| 867 | TraceEvaluation(currentInstr, cur);
|
---|
| 868 | return cur;
|
---|
| 869 | }
|
---|
| 870 | case OpCodes.NumberDataPointsBelowMean: {
|
---|
| 871 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 872 | cur = AggregateApply(cur,
|
---|
| 873 | s => 0,
|
---|
| 874 | v => {
|
---|
| 875 | double mean = Statistics.Mean(v);
|
---|
[17930] | 876 | return v.Map(e => e < mean ? 1.0 : 0.0).Sum();
|
---|
[17830] | 877 | });
|
---|
| 878 | TraceEvaluation(currentInstr, cur);
|
---|
| 879 | return cur;
|
---|
| 880 | }
|
---|
| 881 | case OpCodes.NumberDataPointsBelowMedian: {
|
---|
| 882 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 883 | cur = AggregateApply(cur,
|
---|
| 884 | s => 0,
|
---|
| 885 | v => {
|
---|
| 886 | double median = Statistics.Median(v);
|
---|
[17930] | 887 | return v.Map(e => e < median ? 1.0 : 0.0).Sum();
|
---|
[17830] | 888 | });
|
---|
| 889 | TraceEvaluation(currentInstr, cur);
|
---|
| 890 | return cur;
|
---|
| 891 | }
|
---|
| 892 |
|
---|
| 893 | case OpCodes.ArimaModelCoefficients: {
|
---|
| 894 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 895 | var i = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 896 | var k = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 897 | cur = AggregateApply(cur,
|
---|
| 898 | s => 0,
|
---|
| 899 | v => throw new NotImplementedException(""));
|
---|
| 900 | TraceEvaluation(currentInstr, cur);
|
---|
| 901 | return cur;
|
---|
| 902 | }
|
---|
| 903 | case OpCodes.ContinuousWaveletTransformationCoefficients: {
|
---|
| 904 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 905 | var a = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 906 | var b = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 907 | cur = AggregateApply(cur,
|
---|
| 908 | s => 0,
|
---|
| 909 | v => throw new NotImplementedException(""));
|
---|
| 910 | TraceEvaluation(currentInstr, cur);
|
---|
| 911 | return cur;
|
---|
| 912 | }
|
---|
| 913 | case OpCodes.FastFourierTransformationCoefficient: {
|
---|
| 914 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 915 | var k = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 916 | cur = AggregateApply(cur,
|
---|
| 917 | s => 0,
|
---|
| 918 | v => throw new NotImplementedException(""));
|
---|
| 919 | TraceEvaluation(currentInstr, cur);
|
---|
| 920 | return cur;
|
---|
| 921 | }
|
---|
| 922 | case OpCodes.FirstIndexMax: {
|
---|
| 923 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 924 | cur = AggregateApply(cur,
|
---|
| 925 | s => 0,
|
---|
| 926 | v => (double)v.MaximumIndex() / v.Count);
|
---|
| 927 | TraceEvaluation(currentInstr, cur);
|
---|
| 928 | return cur;
|
---|
| 929 | }
|
---|
| 930 | case OpCodes.FirstIndexMin: {
|
---|
| 931 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 932 | cur = AggregateApply(cur,
|
---|
| 933 | s => 0,
|
---|
| 934 | v => (double)v.MinimumIndex() / v.Count);
|
---|
| 935 | TraceEvaluation(currentInstr, cur);
|
---|
| 936 | return cur;
|
---|
| 937 | }
|
---|
| 938 | case OpCodes.LastIndexMax: {
|
---|
| 939 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 940 | cur = AggregateApply(cur,
|
---|
| 941 | s => 0,
|
---|
| 942 | v => (double)(v.Count - DoubleVector.Build.DenseOfEnumerable(v.Reverse()).MaximumIndex()) / v.Count);
|
---|
| 943 |
|
---|
| 944 | TraceEvaluation(currentInstr, cur);
|
---|
| 945 | return cur;
|
---|
| 946 | }
|
---|
| 947 | case OpCodes.LastIndexMin: {
|
---|
| 948 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 949 | cur = AggregateApply(cur,
|
---|
| 950 | s => 0,
|
---|
| 951 | v => (double)(v.Count - DoubleVector.Build.DenseOfEnumerable(v.Reverse()).MinimumIndex()) / v.Count);
|
---|
| 952 | TraceEvaluation(currentInstr, cur);
|
---|
| 953 | return cur;
|
---|
| 954 | }
|
---|
| 955 | case OpCodes.LongestStrikeAboveMean: {
|
---|
| 956 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 957 | cur = AggregateApply(cur,
|
---|
| 958 | s => 0,
|
---|
| 959 | v => LongestStrikeAbove(v, Statistics.Mean(v)));
|
---|
| 960 | TraceEvaluation(currentInstr, cur);
|
---|
| 961 | return cur;
|
---|
| 962 | }
|
---|
| 963 | case OpCodes.LongestStrikeAboveMedian: {
|
---|
| 964 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 965 | cur = AggregateApply(cur,
|
---|
| 966 | s => 0,
|
---|
| 967 | v => LongestStrikeAbove(v, Statistics.Median(v)));
|
---|
| 968 | TraceEvaluation(currentInstr, cur);
|
---|
| 969 | return cur;
|
---|
| 970 | }
|
---|
| 971 | case OpCodes.LongestStrikeBelowMean: {
|
---|
| 972 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 973 | cur = AggregateApply(cur,
|
---|
| 974 | s => 0,
|
---|
| 975 | v => LongestStrikeBelow(v, Statistics.Mean(v)));
|
---|
| 976 | TraceEvaluation(currentInstr, cur);
|
---|
| 977 | return cur;
|
---|
| 978 | }
|
---|
| 979 | case OpCodes.LongestStrikeBelowMedian: {
|
---|
| 980 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 981 | cur = AggregateApply(cur,
|
---|
| 982 | s => 0,
|
---|
| 983 | v => LongestStrikeBelow(v, Statistics.Median(v)));
|
---|
| 984 | TraceEvaluation(currentInstr, cur);
|
---|
| 985 | return cur;
|
---|
| 986 | }
|
---|
| 987 | case OpCodes.LongestStrikePositive: {
|
---|
| 988 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 989 | cur = AggregateApply(cur,
|
---|
| 990 | s => 0,
|
---|
| 991 | v => LongestStrikeAbove(v, 0));
|
---|
| 992 | TraceEvaluation(currentInstr, cur);
|
---|
| 993 | return cur;
|
---|
| 994 | }
|
---|
| 995 | case OpCodes.LongestStrikeNegative: {
|
---|
| 996 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 997 | cur = AggregateApply(cur,
|
---|
| 998 | s => 0,
|
---|
| 999 | v => LongestStrikeAbove(v, 0));
|
---|
| 1000 | TraceEvaluation(currentInstr, cur);
|
---|
| 1001 | return cur;
|
---|
| 1002 | }
|
---|
| 1003 | case OpCodes.LongestStrikeZero: {
|
---|
| 1004 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 1005 | cur = AggregateApply(cur,
|
---|
| 1006 | s => 0,
|
---|
| 1007 | v => LongestStrikeEqual(v, 0));
|
---|
| 1008 | TraceEvaluation(currentInstr, cur);
|
---|
| 1009 | return cur;
|
---|
| 1010 | }
|
---|
| 1011 | case OpCodes.MeanAbsoluteChange: {
|
---|
| 1012 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 1013 | cur = AggregateApply(cur,
|
---|
| 1014 | s => 0,
|
---|
| 1015 | v => {
|
---|
| 1016 | double sum = 0.0;
|
---|
| 1017 | for (int i = 0; i < v.Count - 1; i++) {
|
---|
| 1018 | sum += Math.Abs(v[i + 1] - v[i]);
|
---|
| 1019 | }
|
---|
| 1020 |
|
---|
| 1021 | return sum / v.Count;
|
---|
| 1022 | });
|
---|
| 1023 | TraceEvaluation(currentInstr, cur);
|
---|
| 1024 | return cur;
|
---|
| 1025 | }
|
---|
| 1026 | case OpCodes.MeanAbsoluteChangeQuantiles: {
|
---|
| 1027 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 1028 | var ql = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 1029 | var qu = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 1030 | cur = AggregateApply(cur,
|
---|
| 1031 | s => 0,
|
---|
| 1032 | v => {
|
---|
| 1033 | var lowerBound = Statistics.Quantile(v, ql.Scalar);
|
---|
| 1034 | var upperBound = Statistics.Quantile(v, qu.Scalar);
|
---|
| 1035 | var inBounds = v.Select(e => e > lowerBound && e < upperBound).ToList();
|
---|
| 1036 | double sum = 0.0;
|
---|
| 1037 | int count = 0;
|
---|
| 1038 | for (int i = 0; i < v.Count - 1; i++) {
|
---|
| 1039 | if (inBounds[i] && inBounds[i + 1]) {
|
---|
| 1040 | sum += Math.Abs(v[i + 1] - v[i]);
|
---|
| 1041 | count++;
|
---|
| 1042 | }
|
---|
| 1043 | }
|
---|
| 1044 |
|
---|
| 1045 | return sum / count;
|
---|
| 1046 | });
|
---|
| 1047 | TraceEvaluation(currentInstr, cur);
|
---|
| 1048 | return cur;
|
---|
| 1049 | }
|
---|
| 1050 | case OpCodes.MeanAutocorrelation: {
|
---|
| 1051 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 1052 | cur = AggregateApply(cur,
|
---|
| 1053 | s => 0,
|
---|
| 1054 | v => {
|
---|
| 1055 | double sum = 0.0;
|
---|
| 1056 | double mean = Statistics.Mean(v);
|
---|
| 1057 | for (int l = 0; l < v.Count; l++) {
|
---|
| 1058 | for (int i = 0; i < v.Count - l; i++) {
|
---|
| 1059 | sum += (v[i] - mean) * (v[i + l] - mean);
|
---|
| 1060 | }
|
---|
| 1061 | }
|
---|
| 1062 |
|
---|
| 1063 | return sum / (v.Count - 1) / Statistics.PopulationVariance(v);
|
---|
| 1064 | });
|
---|
| 1065 | TraceEvaluation(currentInstr, cur);
|
---|
| 1066 | return cur;
|
---|
| 1067 | }
|
---|
| 1068 | case OpCodes.LaggedAutocorrelation: {
|
---|
| 1069 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 1070 | var lVal = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 1071 | cur = AggregateApply(cur,
|
---|
| 1072 | s => 0,
|
---|
| 1073 | v => {
|
---|
| 1074 | double sum = 0.0;
|
---|
[17930] | 1075 | int l = Math.Max((int)Math.Round(lVal.Scalar), 0);
|
---|
[17830] | 1076 | double mean = Statistics.Mean(v);
|
---|
| 1077 | for (int i = 0; i < v.Count - l; i++) {
|
---|
| 1078 | sum += (v[i] - mean) * (v[i + l] - mean);
|
---|
| 1079 | }
|
---|
| 1080 |
|
---|
| 1081 | return sum / Statistics.PopulationVariance(v);
|
---|
| 1082 | });
|
---|
| 1083 | TraceEvaluation(currentInstr, cur);
|
---|
| 1084 | return cur;
|
---|
| 1085 | }
|
---|
| 1086 | case OpCodes.MeanSecondDerivateCentral: {
|
---|
| 1087 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 1088 | cur = AggregateApply(cur,
|
---|
| 1089 | s => 0,
|
---|
| 1090 | v => {
|
---|
| 1091 | double sum = 0.0;
|
---|
| 1092 | for (int i = 1; i < v.Count - 1; i++) {
|
---|
| 1093 | sum += (v[i - 1] - 2 * v[i] + v[i + 1]) / 2;
|
---|
| 1094 | }
|
---|
| 1095 |
|
---|
| 1096 | return sum / (v.Count - 2);
|
---|
| 1097 | });
|
---|
| 1098 | TraceEvaluation(currentInstr, cur);
|
---|
| 1099 | return cur;
|
---|
| 1100 | }
|
---|
| 1101 | case OpCodes.NumberPeaksOfSize: {
|
---|
| 1102 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 1103 | var l = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 1104 | cur = AggregateApply(cur,
|
---|
| 1105 | s => 0,
|
---|
| 1106 | v => CountNumberOfPeaks(v, l.Scalar));
|
---|
| 1107 | TraceEvaluation(currentInstr, cur);
|
---|
| 1108 | return cur;
|
---|
| 1109 | }
|
---|
| 1110 | case OpCodes.LargeNumberOfPeaks: {
|
---|
| 1111 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 1112 | var l = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 1113 | var m = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 1114 | cur = AggregateApply(cur,
|
---|
| 1115 | s => 0,
|
---|
| 1116 | v => CountNumberOfPeaks(v, l.Scalar) > m.Scalar ? 1.0 : 0.0);
|
---|
| 1117 | TraceEvaluation(currentInstr, cur);
|
---|
| 1118 | return cur;
|
---|
| 1119 | }
|
---|
| 1120 | case OpCodes.TimeReversalAsymmetryStatistic: {
|
---|
| 1121 | var cur = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 1122 | var l = Evaluate(dataset, ref row, state, traceDict);
|
---|
| 1123 | cur = AggregateApply(cur,
|
---|
| 1124 | s => 0,
|
---|
| 1125 | v => {
|
---|
[17930] | 1126 | int lag = Math.Max((int)Math.Round(l.Scalar), 0);
|
---|
[17830] | 1127 | double sum = 0.0;
|
---|
| 1128 | for (int i = 0; i < v.Count - 2 * lag; i++) {
|
---|
| 1129 | sum += Math.Pow(v[i + 2 * lag], 2) * v[i + lag] - v[i + lag] * Math.Pow(v[i], 2);
|
---|
| 1130 | }
|
---|
| 1131 |
|
---|
| 1132 | return sum / (v.Count - 2 * lag);
|
---|
| 1133 | });
|
---|
| 1134 | TraceEvaluation(currentInstr, cur);
|
---|
| 1135 | return cur;
|
---|
| 1136 | }
|
---|
| 1137 | #endregion
|
---|
| 1138 |
|
---|
[13248] | 1139 | default:
|
---|
[17455] | 1140 | throw new NotSupportedException($"Unsupported OpCode: {currentInstr.opCode}");
|
---|
[5571] | 1141 | }
|
---|
| 1142 | }
|
---|
[17830] | 1143 |
|
---|
| 1144 | private static int LongestStrikeAbove(DoubleVector v, double threshold) {
|
---|
| 1145 | int longestStrike = 0, currentStrike = 0;
|
---|
| 1146 | for (int i = 0; i < v.Count; i++) {
|
---|
| 1147 | if (v[i] > threshold) {
|
---|
| 1148 | currentStrike++;
|
---|
| 1149 | longestStrike = Math.Max(longestStrike, currentStrike);
|
---|
| 1150 | } else
|
---|
| 1151 | currentStrike = 0;
|
---|
| 1152 | }
|
---|
| 1153 | return longestStrike;
|
---|
| 1154 | }
|
---|
| 1155 | private static int LongestStrikeBelow(DoubleVector v, double threshold) {
|
---|
| 1156 | int longestStrike = 0, currentStrike = 0;
|
---|
| 1157 | for (int i = 0; i < v.Count; i++) {
|
---|
| 1158 | if (v[i] < threshold) {
|
---|
| 1159 | currentStrike++;
|
---|
| 1160 | longestStrike = Math.Max(longestStrike, currentStrike);
|
---|
| 1161 | } else
|
---|
| 1162 | currentStrike = 0;
|
---|
| 1163 | }
|
---|
| 1164 | return longestStrike;
|
---|
| 1165 | }
|
---|
| 1166 |
|
---|
| 1167 | private static int LongestStrikeEqual(DoubleVector v, double value, double epsilon = double.Epsilon) {
|
---|
| 1168 | int longestStrike = 0, currentStrike = 0;
|
---|
| 1169 | for (int i = 0; i < v.Count; i++) {
|
---|
| 1170 | if (v[i].IsAlmost(epsilon)) {
|
---|
| 1171 | currentStrike++;
|
---|
| 1172 | longestStrike = Math.Max(longestStrike, currentStrike);
|
---|
| 1173 | } else
|
---|
| 1174 | currentStrike = 0;
|
---|
| 1175 | }
|
---|
| 1176 | return longestStrike;
|
---|
| 1177 | }
|
---|
| 1178 | private static int CountNumberOfPeaks(DoubleVector v, double heightDifference) {
|
---|
| 1179 | int count = 0;
|
---|
| 1180 | for (int i = 0; i < v.Count; i++) {
|
---|
| 1181 | bool largerThanPrev = i == 0 || v[i] > v[i - 1] + heightDifference;
|
---|
| 1182 | bool largerThanNext = i == v.Count - 1 || v[i] > v[i + 1] + heightDifference;
|
---|
| 1183 | if (largerThanPrev && largerThanNext)
|
---|
| 1184 | count++;
|
---|
| 1185 | }
|
---|
| 1186 | return count;
|
---|
| 1187 | }
|
---|
[5571] | 1188 | }
|
---|
[13248] | 1189 | } |
---|