1 | #region License Information
|
---|
2 |
|
---|
3 | /* HeuristicLab
|
---|
4 | * Copyright (C) 2002-2018 Heuristic and Evolutionary Algorithms Laboratory (HEAL)
|
---|
5 | *
|
---|
6 | * This file is part of HeuristicLab.
|
---|
7 | *
|
---|
8 | * HeuristicLab is free software: you can redistribute it and/or modify
|
---|
9 | * it under the terms of the GNU General Public License as published by
|
---|
10 | * the Free Software Foundation, either version 3 of the License, or
|
---|
11 | * (at your option) any later version.
|
---|
12 | *
|
---|
13 | * HeuristicLab is distributed in the hope that it will be useful,
|
---|
14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
---|
15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
---|
16 | * GNU General Public License for more details.
|
---|
17 | *
|
---|
18 | * You should have received a copy of the GNU General Public License
|
---|
19 | * along with HeuristicLab. If not, see <http://www.gnu.org/licenses/>.
|
---|
20 | */
|
---|
21 |
|
---|
22 | #endregion
|
---|
23 |
|
---|
24 | using System;
|
---|
25 | using System.Collections;
|
---|
26 | using System.Collections.Generic;
|
---|
27 | using System.Linq;
|
---|
28 | using HeuristicLab.Common;
|
---|
29 | using HeuristicLab.Core;
|
---|
30 | using HeuristicLab.Data;
|
---|
31 | using HeuristicLab.Parameters;
|
---|
32 | using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
|
---|
33 | using HeuristicLab.Random;
|
---|
34 |
|
---|
35 | namespace HeuristicLab.Problems.DataAnalysis {
|
---|
36 | [StorableClass]
|
---|
37 | [Item("RegressionSolution Impacts Calculator", "Calculation of the impacts of input variables for any regression solution")]
|
---|
38 | public sealed class RegressionSolutionVariableImpactsCalculator : ParameterizedNamedItem {
|
---|
39 | public enum ReplacementMethodEnum {
|
---|
40 | Median,
|
---|
41 | Average,
|
---|
42 | Shuffle,
|
---|
43 | Noise
|
---|
44 | }
|
---|
45 | public enum FactorReplacementMethodEnum {
|
---|
46 | Best,
|
---|
47 | Mode,
|
---|
48 | Shuffle
|
---|
49 | }
|
---|
50 | public enum DataPartitionEnum {
|
---|
51 | Training,
|
---|
52 | Test,
|
---|
53 | All
|
---|
54 | }
|
---|
55 |
|
---|
56 | private const string ReplacementParameterName = "Replacement Method";
|
---|
57 | private const string FactorReplacementParameterName = "Factor Replacement Method";
|
---|
58 | private const string DataPartitionParameterName = "DataPartition";
|
---|
59 |
|
---|
60 | public IFixedValueParameter<EnumValue<ReplacementMethodEnum>> ReplacementParameter
|
---|
61 | {
|
---|
62 | get { return (IFixedValueParameter<EnumValue<ReplacementMethodEnum>>)Parameters[ReplacementParameterName]; }
|
---|
63 | }
|
---|
64 | public IFixedValueParameter<EnumValue<FactorReplacementMethodEnum>> FactorReplacementParameter
|
---|
65 | {
|
---|
66 | get { return (IFixedValueParameter<EnumValue<FactorReplacementMethodEnum>>)Parameters[FactorReplacementParameterName]; }
|
---|
67 | }
|
---|
68 | public IFixedValueParameter<EnumValue<DataPartitionEnum>> DataPartitionParameter
|
---|
69 | {
|
---|
70 | get { return (IFixedValueParameter<EnumValue<DataPartitionEnum>>)Parameters[DataPartitionParameterName]; }
|
---|
71 | }
|
---|
72 |
|
---|
73 | public ReplacementMethodEnum ReplacementMethod
|
---|
74 | {
|
---|
75 | get { return ReplacementParameter.Value.Value; }
|
---|
76 | set { ReplacementParameter.Value.Value = value; }
|
---|
77 | }
|
---|
78 | public FactorReplacementMethodEnum FactorReplacementMethod
|
---|
79 | {
|
---|
80 | get { return FactorReplacementParameter.Value.Value; }
|
---|
81 | set { FactorReplacementParameter.Value.Value = value; }
|
---|
82 | }
|
---|
83 | public DataPartitionEnum DataPartition
|
---|
84 | {
|
---|
85 | get { return DataPartitionParameter.Value.Value; }
|
---|
86 | set { DataPartitionParameter.Value.Value = value; }
|
---|
87 | }
|
---|
88 |
|
---|
89 |
|
---|
90 | [StorableConstructor]
|
---|
91 | private RegressionSolutionVariableImpactsCalculator(bool deserializing) : base(deserializing) { }
|
---|
92 | private RegressionSolutionVariableImpactsCalculator(RegressionSolutionVariableImpactsCalculator original, Cloner cloner)
|
---|
93 | : base(original, cloner) { }
|
---|
94 | public override IDeepCloneable Clone(Cloner cloner) {
|
---|
95 | return new RegressionSolutionVariableImpactsCalculator(this, cloner);
|
---|
96 | }
|
---|
97 |
|
---|
98 | public RegressionSolutionVariableImpactsCalculator()
|
---|
99 | : base() {
|
---|
100 | Parameters.Add(new FixedValueParameter<EnumValue<ReplacementMethodEnum>>(ReplacementParameterName, "The replacement method for variables during impact calculation.", new EnumValue<ReplacementMethodEnum>(ReplacementMethodEnum.Median)));
|
---|
101 | Parameters.Add(new FixedValueParameter<EnumValue<FactorReplacementMethodEnum>>(FactorReplacementParameterName, "The replacement method for factor variables during impact calculation.", new EnumValue<FactorReplacementMethodEnum>(FactorReplacementMethodEnum.Best)));
|
---|
102 | Parameters.Add(new FixedValueParameter<EnumValue<DataPartitionEnum>>(DataPartitionParameterName, "The data partition on which the impacts are calculated.", new EnumValue<DataPartitionEnum>(DataPartitionEnum.Training)));
|
---|
103 | }
|
---|
104 |
|
---|
105 | //mkommend: annoying name clash with static method, open to better naming suggestions
|
---|
106 | public IEnumerable<Tuple<string, double>> Calculate(IRegressionSolution solution) {
|
---|
107 | return CalculateImpacts(solution, ReplacementMethod, FactorReplacementMethod, DataPartition);
|
---|
108 | }
|
---|
109 |
|
---|
110 | public static IEnumerable<Tuple<string, double>> CalculateImpacts(
|
---|
111 | IRegressionSolution solution,
|
---|
112 | ReplacementMethodEnum replacementMethod = ReplacementMethodEnum.Shuffle,
|
---|
113 | FactorReplacementMethodEnum factorReplacementMethod = FactorReplacementMethodEnum.Best,
|
---|
114 | DataPartitionEnum data = DataPartitionEnum.Training,
|
---|
115 | Func<double, string, bool> progressCallback = null) {
|
---|
116 | return CalculateImpacts(solution.Model, solution.ProblemData, solution.EstimatedValues, replacementMethod, factorReplacementMethod, data, progressCallback);
|
---|
117 | }
|
---|
118 |
|
---|
119 | public static IEnumerable<Tuple<string, double>> CalculateImpacts(
|
---|
120 | IRegressionModel model,
|
---|
121 | IRegressionProblemData problemData,
|
---|
122 | IEnumerable<double> estimatedValues,
|
---|
123 | ReplacementMethodEnum replacementMethod = ReplacementMethodEnum.Shuffle,
|
---|
124 | FactorReplacementMethodEnum factorReplacementMethod = FactorReplacementMethodEnum.Best,
|
---|
125 | DataPartitionEnum data = DataPartitionEnum.Training,
|
---|
126 | Func<double, string, bool> progressCallback = null) {
|
---|
127 | IEnumerable<int> rows;
|
---|
128 |
|
---|
129 | switch (data) {
|
---|
130 | case DataPartitionEnum.All:
|
---|
131 | rows = problemData.AllIndices;
|
---|
132 | break;
|
---|
133 | case DataPartitionEnum.Test:
|
---|
134 | rows = problemData.TestIndices;
|
---|
135 | break;
|
---|
136 | case DataPartitionEnum.Training:
|
---|
137 | rows = problemData.TrainingIndices;
|
---|
138 | break;
|
---|
139 | default:
|
---|
140 | throw new NotSupportedException("DataPartition not supported");
|
---|
141 | }
|
---|
142 |
|
---|
143 | return CalculateImpacts(model, problemData, estimatedValues, rows, replacementMethod, factorReplacementMethod, progressCallback);
|
---|
144 | }
|
---|
145 |
|
---|
146 | public static IEnumerable<Tuple<string, double>> CalculateImpacts(
|
---|
147 | IRegressionModel model,
|
---|
148 | IRegressionProblemData problemData,
|
---|
149 | IEnumerable<double> estimatedValues,
|
---|
150 | IEnumerable<int> rows,
|
---|
151 | ReplacementMethodEnum replacementMethod = ReplacementMethodEnum.Shuffle,
|
---|
152 | FactorReplacementMethodEnum factorReplacementMethod = FactorReplacementMethodEnum.Best,
|
---|
153 | Func<double, string, bool> progressCallback = null) {
|
---|
154 |
|
---|
155 | IEnumerable<double> targetValues;
|
---|
156 | double originalCalculatorValue = -1;
|
---|
157 |
|
---|
158 | PrepareData(rows, problemData, estimatedValues, out targetValues, out originalCalculatorValue);
|
---|
159 |
|
---|
160 | var impacts = new Dictionary<string, double>();
|
---|
161 | var inputvariables = new HashSet<string>(problemData.AllowedInputVariables.Union(model.VariablesUsedForPrediction));
|
---|
162 | var allowedInputVariables = problemData.Dataset.VariableNames.Where(v => inputvariables.Contains(v)).ToList();
|
---|
163 |
|
---|
164 | int curIdx = 0;
|
---|
165 | int count = allowedInputVariables.Count(v => problemData.Dataset.VariableHasType<double>(v) || problemData.Dataset.VariableHasType<string>(v));
|
---|
166 |
|
---|
167 | foreach (var inputVariable in allowedInputVariables) {
|
---|
168 | //Report the current progress in percent. If the callback returns true, it means the execution shall be stopped
|
---|
169 | if (progressCallback != null) {
|
---|
170 | curIdx++;
|
---|
171 | if (progressCallback((double)curIdx / count, string.Format("Calculating impact for variable {0} ({1} of {2})", inputVariable, curIdx, count))) { return null; }
|
---|
172 | }
|
---|
173 | impacts[inputVariable] = CalculateImpact(inputVariable, model, problemData.Dataset, rows, targetValues, originalCalculatorValue, replacementMethod, factorReplacementMethod);
|
---|
174 | }
|
---|
175 |
|
---|
176 | return impacts.OrderByDescending(i => i.Value).Select(i => Tuple.Create(i.Key, i.Value));
|
---|
177 | }
|
---|
178 |
|
---|
179 | public static double CalculateImpact(string variableName,
|
---|
180 | IRegressionSolution solution,
|
---|
181 | IEnumerable<int> rows,
|
---|
182 | IEnumerable<double> targetValues,
|
---|
183 | double originalValue,
|
---|
184 | ReplacementMethodEnum replacementMethod = ReplacementMethodEnum.Shuffle,
|
---|
185 | FactorReplacementMethodEnum factorReplacementMethod = FactorReplacementMethodEnum.Best,
|
---|
186 | DataPartitionEnum data = DataPartitionEnum.Training) {
|
---|
187 | return CalculateImpact(variableName, solution.Model, solution.ProblemData.Dataset, rows, targetValues, originalValue, replacementMethod, factorReplacementMethod);
|
---|
188 | }
|
---|
189 |
|
---|
190 | public static double CalculateImpact(string variableName,
|
---|
191 | IRegressionModel model,
|
---|
192 | IDataset dataset,
|
---|
193 | IEnumerable<int> rows,
|
---|
194 | IEnumerable<double> targetValues,
|
---|
195 | double originalValue,
|
---|
196 | ReplacementMethodEnum replacementMethod = ReplacementMethodEnum.Shuffle,
|
---|
197 | FactorReplacementMethodEnum factorReplacementMethod = FactorReplacementMethodEnum.Best) {
|
---|
198 |
|
---|
199 | double impact = 0;
|
---|
200 | var modifiableDataset = ((Dataset)(dataset).Clone()).ToModifiable();
|
---|
201 |
|
---|
202 | // calculate impacts for double variables
|
---|
203 | if (dataset.VariableHasType<double>(variableName)) {
|
---|
204 | impact = CalculateImpactForDouble(variableName, model, modifiableDataset, rows, targetValues, originalValue, replacementMethod);
|
---|
205 | } else if (dataset.VariableHasType<string>(variableName)) {
|
---|
206 | impact = CalculateImpactForString(variableName, model, dataset, modifiableDataset, rows, targetValues, originalValue, factorReplacementMethod);
|
---|
207 | } else {
|
---|
208 | throw new NotSupportedException("Variable not supported");
|
---|
209 | }
|
---|
210 | return impact;
|
---|
211 | }
|
---|
212 |
|
---|
213 | private static void PrepareData(IEnumerable<int> rows,
|
---|
214 | IRegressionProblemData problemData,
|
---|
215 | IEnumerable<double> estimatedValues,
|
---|
216 | out IEnumerable<double> targetValues,
|
---|
217 | out double originalValue) {
|
---|
218 | OnlineCalculatorError error;
|
---|
219 |
|
---|
220 | var targetVariableValueList = problemData.TargetVariableValues.ToList();
|
---|
221 | targetValues = rows.Select(v => targetVariableValueList.ElementAt(v));
|
---|
222 | var estimatedValuesPartition = rows.Select(v => estimatedValues.ElementAt(v));
|
---|
223 | originalValue = CalculateValue(targetValues, estimatedValuesPartition, out error);
|
---|
224 |
|
---|
225 | if (error != OnlineCalculatorError.None) throw new InvalidOperationException("Error during calculation.");
|
---|
226 | }
|
---|
227 |
|
---|
228 | private static double CalculateImpactForDouble(string variableName,
|
---|
229 | IRegressionModel model,
|
---|
230 | ModifiableDataset modifiableDataset,
|
---|
231 | IEnumerable<int> rows,
|
---|
232 | IEnumerable<double> targetValues,
|
---|
233 | double originalValue,
|
---|
234 | ReplacementMethodEnum replacementMethod) {
|
---|
235 | OnlineCalculatorError error;
|
---|
236 | var newEstimates = EvaluateModelWithReplacedVariable(model, variableName, modifiableDataset, rows, replacementMethod);
|
---|
237 | var newValue = CalculateValue(targetValues, newEstimates, out error);
|
---|
238 | if (error != OnlineCalculatorError.None) { throw new InvalidOperationException("Error during calculation with replaced inputs."); }
|
---|
239 | return originalValue - newValue;
|
---|
240 | }
|
---|
241 |
|
---|
242 | private static double CalculateImpactForString(string variableName,
|
---|
243 | IRegressionModel model,
|
---|
244 | IDataset problemData,
|
---|
245 | ModifiableDataset modifiableDataset,
|
---|
246 | IEnumerable<int> rows,
|
---|
247 | IEnumerable<double> targetValues,
|
---|
248 | double originalValue,
|
---|
249 | FactorReplacementMethodEnum factorReplacementMethod) {
|
---|
250 |
|
---|
251 | OnlineCalculatorError error;
|
---|
252 | if (factorReplacementMethod == FactorReplacementMethodEnum.Best) {
|
---|
253 | // try replacing with all possible values and find the best replacement value
|
---|
254 | var smallestImpact = double.PositiveInfinity;
|
---|
255 | foreach (var repl in problemData.GetStringValues(variableName, rows).Distinct()) {
|
---|
256 | var originalValues = modifiableDataset.GetReadOnlyStringValues(variableName).ToList();
|
---|
257 | var newEstimates = EvaluateModelWithReplacedVariable(originalValues, model, variableName, modifiableDataset, rows, Enumerable.Repeat(repl, problemData.Rows).ToList());
|
---|
258 | var newValue = CalculateValue(targetValues, newEstimates, out error);
|
---|
259 | if (error != OnlineCalculatorError.None) throw new InvalidOperationException("Error during calculation with replaced inputs.");
|
---|
260 |
|
---|
261 | var curImpact = originalValue - newValue;
|
---|
262 | if (curImpact < smallestImpact) smallestImpact = curImpact;
|
---|
263 | }
|
---|
264 | return smallestImpact;
|
---|
265 | } else {
|
---|
266 | // for replacement methods shuffle and mode
|
---|
267 | // calculate impacts for factor variables
|
---|
268 | var newEstimates = EvaluateModelWithReplacedVariable(model, variableName, modifiableDataset, rows, factorReplacementMethod);
|
---|
269 | var newValue = CalculateValue(targetValues, newEstimates, out error);
|
---|
270 | if (error != OnlineCalculatorError.None) throw new InvalidOperationException("Error during calculation with replaced inputs.");
|
---|
271 |
|
---|
272 | return originalValue - newValue;
|
---|
273 | }
|
---|
274 | }
|
---|
275 |
|
---|
276 | private static IEnumerable<double> EvaluateModelWithReplacedVariable(IRegressionModel model, string variable, ModifiableDataset dataset, IEnumerable<int> rows, ReplacementMethodEnum replacement = ReplacementMethodEnum.Shuffle) {
|
---|
277 | var originalValues = dataset.GetReadOnlyDoubleValues(variable).ToList();
|
---|
278 | double replacementValue;
|
---|
279 | List<double> replacementValues;
|
---|
280 | IRandom rand;
|
---|
281 |
|
---|
282 | switch (replacement) {
|
---|
283 | case ReplacementMethodEnum.Median:
|
---|
284 | replacementValue = rows.Select(r => originalValues[r]).Median();
|
---|
285 | replacementValues = Enumerable.Repeat(replacementValue, dataset.Rows).ToList();
|
---|
286 | break;
|
---|
287 | case ReplacementMethodEnum.Average:
|
---|
288 | replacementValue = rows.Select(r => originalValues[r]).Average();
|
---|
289 | replacementValues = Enumerable.Repeat(replacementValue, dataset.Rows).ToList();
|
---|
290 | break;
|
---|
291 | case ReplacementMethodEnum.Shuffle:
|
---|
292 | // new var has same empirical distribution but the relation to y is broken
|
---|
293 | rand = new FastRandom(31415);
|
---|
294 | // prepare a complete column for the dataset
|
---|
295 | replacementValues = Enumerable.Repeat(double.NaN, dataset.Rows).ToList();
|
---|
296 | // shuffle only the selected rows
|
---|
297 | var shuffledValues = rows.Select(r => originalValues[r]).Shuffle(rand).ToList();
|
---|
298 | int i = 0;
|
---|
299 | // update column values
|
---|
300 | foreach (var r in rows) {
|
---|
301 | replacementValues[r] = shuffledValues[i++];
|
---|
302 | }
|
---|
303 | break;
|
---|
304 | case ReplacementMethodEnum.Noise:
|
---|
305 | var avg = rows.Select(r => originalValues[r]).Average();
|
---|
306 | var stdDev = rows.Select(r => originalValues[r]).StandardDeviation();
|
---|
307 | rand = new FastRandom(31415);
|
---|
308 | // prepare a complete column for the dataset
|
---|
309 | replacementValues = Enumerable.Repeat(double.NaN, dataset.Rows).ToList();
|
---|
310 | // update column values
|
---|
311 | foreach (var r in rows) {
|
---|
312 | replacementValues[r] = NormalDistributedRandom.NextDouble(rand, avg, stdDev);
|
---|
313 | }
|
---|
314 | break;
|
---|
315 |
|
---|
316 | default:
|
---|
317 | throw new ArgumentException(string.Format("ReplacementMethod {0} cannot be handled.", replacement));
|
---|
318 | }
|
---|
319 |
|
---|
320 | return EvaluateModelWithReplacedVariable(originalValues, model, variable, dataset, rows, replacementValues);
|
---|
321 | }
|
---|
322 |
|
---|
323 | private static IEnumerable<double> EvaluateModelWithReplacedVariable(
|
---|
324 | IRegressionModel model, string variable, ModifiableDataset dataset,
|
---|
325 | IEnumerable<int> rows,
|
---|
326 | FactorReplacementMethodEnum replacement = FactorReplacementMethodEnum.Best) {
|
---|
327 | var originalValues = dataset.GetReadOnlyStringValues(variable).ToList();
|
---|
328 | List<string> replacementValues;
|
---|
329 | IRandom rand;
|
---|
330 |
|
---|
331 | switch (replacement) {
|
---|
332 | case FactorReplacementMethodEnum.Mode:
|
---|
333 | var mostCommonValue = rows.Select(r => originalValues[r])
|
---|
334 | .GroupBy(v => v)
|
---|
335 | .OrderByDescending(g => g.Count())
|
---|
336 | .First().Key;
|
---|
337 | replacementValues = Enumerable.Repeat(mostCommonValue, dataset.Rows).ToList();
|
---|
338 | break;
|
---|
339 | case FactorReplacementMethodEnum.Shuffle:
|
---|
340 | // new var has same empirical distribution but the relation to y is broken
|
---|
341 | rand = new FastRandom(31415);
|
---|
342 | // prepare a complete column for the dataset
|
---|
343 | replacementValues = Enumerable.Repeat(string.Empty, dataset.Rows).ToList();
|
---|
344 | // shuffle only the selected rows
|
---|
345 | var shuffledValues = rows.Select(r => originalValues[r]).Shuffle(rand).ToList();
|
---|
346 | int i = 0;
|
---|
347 | // update column values
|
---|
348 | foreach (var r in rows) {
|
---|
349 | replacementValues[r] = shuffledValues[i++];
|
---|
350 | }
|
---|
351 | break;
|
---|
352 | default:
|
---|
353 | throw new ArgumentException(string.Format("FactorReplacementMethod {0} cannot be handled.", replacement));
|
---|
354 | }
|
---|
355 |
|
---|
356 | return EvaluateModelWithReplacedVariable(originalValues, model, variable, dataset, rows, replacementValues);
|
---|
357 | }
|
---|
358 |
|
---|
359 | private static IEnumerable<double> EvaluateModelWithReplacedVariable(IList originalValues, IRegressionModel model, string variable,
|
---|
360 | ModifiableDataset dataset, IEnumerable<int> rows, IList replacementValues) {
|
---|
361 | dataset.ReplaceVariable(variable, replacementValues);
|
---|
362 | //mkommend: ToList is used on purpose to avoid lazy evaluation that could result in wrong estimates due to variable replacements
|
---|
363 | var estimates = model.GetEstimatedValues(dataset, rows).ToList();
|
---|
364 | dataset.ReplaceVariable(variable, originalValues);
|
---|
365 |
|
---|
366 | return estimates;
|
---|
367 | }
|
---|
368 |
|
---|
369 | private static double CalculateValue(IEnumerable<double> originalValues, IEnumerable<double> estimatedValues, out OnlineCalculatorError errorState) {
|
---|
370 | IEnumerator<double> firstEnumerator = originalValues.GetEnumerator();
|
---|
371 | IEnumerator<double> secondEnumerator = estimatedValues.GetEnumerator();
|
---|
372 | var calculator = new OnlinePearsonsRSquaredCalculator();
|
---|
373 |
|
---|
374 | // always move forward both enumerators (do not use short-circuit evaluation!)
|
---|
375 | while (firstEnumerator.MoveNext() & secondEnumerator.MoveNext()) {
|
---|
376 | double original = firstEnumerator.Current;
|
---|
377 | double estimated = secondEnumerator.Current;
|
---|
378 | calculator.Add(original, estimated);
|
---|
379 | if (calculator.ErrorState != OnlineCalculatorError.None) break;
|
---|
380 | }
|
---|
381 |
|
---|
382 | // check if both enumerators are at the end to make sure both enumerations have the same length
|
---|
383 | if (calculator.ErrorState == OnlineCalculatorError.None &&
|
---|
384 | (secondEnumerator.MoveNext() || firstEnumerator.MoveNext())) {
|
---|
385 | throw new ArgumentException("Number of elements in first and second enumeration doesn't match.");
|
---|
386 | } else {
|
---|
387 | errorState = calculator.ErrorState;
|
---|
388 | return calculator.Value;
|
---|
389 | }
|
---|
390 | }
|
---|
391 | }
|
---|
392 | }
|
---|