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 | private static IOnlineCalculator calculator = new OnlinePearsonsRSquaredCalculator();
|
---|
40 |
|
---|
41 | public enum ReplacementMethodEnum {
|
---|
42 | Median,
|
---|
43 | Average,
|
---|
44 | Shuffle,
|
---|
45 | Noise
|
---|
46 | }
|
---|
47 | public enum FactorReplacementMethodEnum {
|
---|
48 | Best,
|
---|
49 | Mode,
|
---|
50 | Shuffle
|
---|
51 | }
|
---|
52 | public enum DataPartitionEnum {
|
---|
53 | Training,
|
---|
54 | Test,
|
---|
55 | All
|
---|
56 | }
|
---|
57 |
|
---|
58 | private const string ReplacementParameterName = "Replacement Method";
|
---|
59 | private const string FactorReplacementParameterName = "Factor Replacement Method";
|
---|
60 | private const string DataPartitionParameterName = "DataPartition";
|
---|
61 |
|
---|
62 | public IFixedValueParameter<EnumValue<ReplacementMethodEnum>> ReplacementParameter
|
---|
63 | {
|
---|
64 | get { return (IFixedValueParameter<EnumValue<ReplacementMethodEnum>>)Parameters[ReplacementParameterName]; }
|
---|
65 | }
|
---|
66 | public IFixedValueParameter<EnumValue<FactorReplacementMethodEnum>> FactorReplacementParameter
|
---|
67 | {
|
---|
68 | get { return (IFixedValueParameter<EnumValue<FactorReplacementMethodEnum>>)Parameters[FactorReplacementParameterName]; }
|
---|
69 | }
|
---|
70 | public IFixedValueParameter<EnumValue<DataPartitionEnum>> DataPartitionParameter
|
---|
71 | {
|
---|
72 | get { return (IFixedValueParameter<EnumValue<DataPartitionEnum>>)Parameters[DataPartitionParameterName]; }
|
---|
73 | }
|
---|
74 |
|
---|
75 | public ReplacementMethodEnum ReplacementMethod
|
---|
76 | {
|
---|
77 | get { return ReplacementParameter.Value.Value; }
|
---|
78 | set { ReplacementParameter.Value.Value = value; }
|
---|
79 | }
|
---|
80 | public FactorReplacementMethodEnum FactorReplacementMethod
|
---|
81 | {
|
---|
82 | get { return FactorReplacementParameter.Value.Value; }
|
---|
83 | set { FactorReplacementParameter.Value.Value = value; }
|
---|
84 | }
|
---|
85 | public DataPartitionEnum DataPartition
|
---|
86 | {
|
---|
87 | get { return DataPartitionParameter.Value.Value; }
|
---|
88 | set { DataPartitionParameter.Value.Value = value; }
|
---|
89 | }
|
---|
90 |
|
---|
91 |
|
---|
92 | [StorableConstructor]
|
---|
93 | private RegressionSolutionVariableImpactsCalculator(bool deserializing) : base(deserializing) { }
|
---|
94 | private RegressionSolutionVariableImpactsCalculator(RegressionSolutionVariableImpactsCalculator original, Cloner cloner)
|
---|
95 | : base(original, cloner) { }
|
---|
96 | public override IDeepCloneable Clone(Cloner cloner) {
|
---|
97 | return new RegressionSolutionVariableImpactsCalculator(this, cloner);
|
---|
98 | }
|
---|
99 |
|
---|
100 | public RegressionSolutionVariableImpactsCalculator()
|
---|
101 | : base() {
|
---|
102 | Parameters.Add(new FixedValueParameter<EnumValue<ReplacementMethodEnum>>(ReplacementParameterName, "The replacement method for variables during impact calculation.", new EnumValue<ReplacementMethodEnum>(ReplacementMethodEnum.Median)));
|
---|
103 | Parameters.Add(new FixedValueParameter<EnumValue<FactorReplacementMethodEnum>>(FactorReplacementParameterName, "The replacement method for factor variables during impact calculation.", new EnumValue<FactorReplacementMethodEnum>(FactorReplacementMethodEnum.Best)));
|
---|
104 | Parameters.Add(new FixedValueParameter<EnumValue<DataPartitionEnum>>(DataPartitionParameterName, "The data partition on which the impacts are calculated.", new EnumValue<DataPartitionEnum>(DataPartitionEnum.Training)));
|
---|
105 | }
|
---|
106 |
|
---|
107 | //mkommend: annoying name clash with static method, open to better naming suggestions
|
---|
108 | public IEnumerable<Tuple<string, double>> Calculate(IRegressionSolution solution) {
|
---|
109 | return CalculateImpacts(solution, DataPartition, ReplacementMethod);
|
---|
110 | }
|
---|
111 |
|
---|
112 | public static IEnumerable<Tuple<string, double>> CalculateImpacts(
|
---|
113 | IRegressionSolution solution,
|
---|
114 | DataPartitionEnum data = DataPartitionEnum.Training,
|
---|
115 | ReplacementMethodEnum replacementMethod = ReplacementMethodEnum.Median,
|
---|
116 | FactorReplacementMethodEnum factorReplacementMethod = FactorReplacementMethodEnum.Best,
|
---|
117 | Func<double, string, bool> progressCallback = null) {
|
---|
118 | return CalculateImpacts(solution.Model, solution.ProblemData, solution.EstimatedValues, data, replacementMethod, factorReplacementMethod, progressCallback);
|
---|
119 | }
|
---|
120 |
|
---|
121 | public static IEnumerable<Tuple<string, double>> CalculateImpacts(
|
---|
122 | IRegressionModel model,
|
---|
123 | IRegressionProblemData problemData,
|
---|
124 | IEnumerable<double> estimatedValues,
|
---|
125 | DataPartitionEnum data = DataPartitionEnum.Training,
|
---|
126 | ReplacementMethodEnum replacementMethod = ReplacementMethodEnum.Median,
|
---|
127 | FactorReplacementMethodEnum factorReplacementMethod = FactorReplacementMethodEnum.Best,
|
---|
128 | Func<double, string, bool> progressCallback = null) {
|
---|
129 | IEnumerable<int> rows;
|
---|
130 |
|
---|
131 | switch (data) {
|
---|
132 | case DataPartitionEnum.All:
|
---|
133 | rows = problemData.AllIndices;
|
---|
134 | break;
|
---|
135 | case DataPartitionEnum.Test:
|
---|
136 | rows = problemData.TestIndices;
|
---|
137 | break;
|
---|
138 | case DataPartitionEnum.Training:
|
---|
139 | rows = problemData.TrainingIndices;
|
---|
140 | break;
|
---|
141 | default:
|
---|
142 | throw new NotSupportedException("DataPartition not supported");
|
---|
143 | }
|
---|
144 |
|
---|
145 | return CalculateImpacts(model, problemData, estimatedValues, rows, replacementMethod, factorReplacementMethod, progressCallback);
|
---|
146 | }
|
---|
147 |
|
---|
148 | public static IEnumerable<Tuple<string, double>> CalculateImpacts(
|
---|
149 | IRegressionModel model,
|
---|
150 | IRegressionProblemData problemData,
|
---|
151 | IEnumerable<double> estimatedValues,
|
---|
152 | IEnumerable<int> rows,
|
---|
153 | ReplacementMethodEnum replacementMethod = ReplacementMethodEnum.Median,
|
---|
154 | FactorReplacementMethodEnum factorReplacementMethod = FactorReplacementMethodEnum.Best,
|
---|
155 | Func<double, string, bool> progressCallback = null) {
|
---|
156 |
|
---|
157 | IEnumerable<double> targetValues;
|
---|
158 | double originalValue = -1;
|
---|
159 |
|
---|
160 | PrepareData(rows, problemData, estimatedValues, out targetValues, out originalValue);
|
---|
161 |
|
---|
162 | var impacts = new Dictionary<string, double>();
|
---|
163 | var inputvariables = new HashSet<string>(problemData.AllowedInputVariables.Union(model.VariablesUsedForPrediction));
|
---|
164 | var allowedInputVariables = problemData.Dataset.VariableNames.Where(v => inputvariables.Contains(v)).ToList();
|
---|
165 |
|
---|
166 | int curIdx = 0;
|
---|
167 | int count = allowedInputVariables.Count();
|
---|
168 |
|
---|
169 | foreach (var inputVariable in allowedInputVariables) {
|
---|
170 | //Report the current progress in percent. If the callback returns true, it means the execution shall be stopped
|
---|
171 | if (progressCallback != null) {
|
---|
172 | curIdx++;
|
---|
173 | if (progressCallback((double)curIdx / count, string.Format("Calculating impact for variable {0} ({1} of {2})", inputVariable, curIdx, count))) { return null; }
|
---|
174 | }
|
---|
175 | impacts[inputVariable] = CalculateImpact(inputVariable, model, problemData.Dataset, rows, targetValues, originalValue, replacementMethod, factorReplacementMethod);
|
---|
176 | }
|
---|
177 |
|
---|
178 | return impacts.OrderByDescending(i => i.Value).Select(i => Tuple.Create(i.Key, i.Value));
|
---|
179 | }
|
---|
180 |
|
---|
181 | public static double CalculateImpact(string variableName,
|
---|
182 | IRegressionSolution solution,
|
---|
183 | IEnumerable<int> rows,
|
---|
184 | IEnumerable<double> targetValues,
|
---|
185 | double originalValue,
|
---|
186 | DataPartitionEnum data = DataPartitionEnum.Training,
|
---|
187 | ReplacementMethodEnum replacementMethod = ReplacementMethodEnum.Median,
|
---|
188 | FactorReplacementMethodEnum factorReplacementMethod = FactorReplacementMethodEnum.Best) {
|
---|
189 | return CalculateImpact(variableName, solution.Model, solution.ProblemData.Dataset, rows, targetValues, originalValue, replacementMethod, factorReplacementMethod);
|
---|
190 | }
|
---|
191 |
|
---|
192 | public static double CalculateImpact(string variableName,
|
---|
193 | IRegressionModel model,
|
---|
194 | IDataset dataset,
|
---|
195 | IEnumerable<int> rows,
|
---|
196 | IEnumerable<double> targetValues,
|
---|
197 | double originalValue,
|
---|
198 | ReplacementMethodEnum replacementMethod = ReplacementMethodEnum.Median,
|
---|
199 | FactorReplacementMethodEnum factorReplacementMethod = FactorReplacementMethodEnum.Best) {
|
---|
200 |
|
---|
201 | double impact = 0;
|
---|
202 | var modifiableDataset = ((Dataset)(dataset).Clone()).ToModifiable();
|
---|
203 |
|
---|
204 | // calculate impacts for double variables
|
---|
205 | if (dataset.VariableHasType<double>(variableName)) {
|
---|
206 | impact = CalculateImpactForDouble(variableName, model, modifiableDataset, rows, targetValues, originalValue, replacementMethod);
|
---|
207 | } else if (dataset.VariableHasType<string>(variableName)) {
|
---|
208 | impact = CalculateImpactForString(variableName, model, dataset, modifiableDataset, rows, targetValues, originalValue, factorReplacementMethod);
|
---|
209 | } else {
|
---|
210 | throw new NotSupportedException("Variable not supported");
|
---|
211 | }
|
---|
212 | return impact;
|
---|
213 | }
|
---|
214 |
|
---|
215 | private static void PrepareData(IEnumerable<int> rows,
|
---|
216 | IRegressionProblemData problemData,
|
---|
217 | IEnumerable<double> estimatedValues,
|
---|
218 | out IEnumerable<double> targetValues,
|
---|
219 | out double originalValue) {
|
---|
220 | OnlineCalculatorError error;
|
---|
221 |
|
---|
222 | var targetVariableValueList = problemData.TargetVariableValues.ToList();
|
---|
223 | targetValues = rows.Select(v => targetVariableValueList.ElementAt(v));
|
---|
224 | var estimatedValuesPartition = rows.Select(v => estimatedValues.ElementAt(v));
|
---|
225 | originalValue = CalculateValue(targetValues, estimatedValuesPartition, out error);
|
---|
226 |
|
---|
227 | if (error != OnlineCalculatorError.None) throw new InvalidOperationException("Error during calculation.");
|
---|
228 | }
|
---|
229 |
|
---|
230 | private static double CalculateImpactForDouble(string variableName,
|
---|
231 | IRegressionModel model,
|
---|
232 | ModifiableDataset modifiableDataset,
|
---|
233 | IEnumerable<int> rows,
|
---|
234 | IEnumerable<double> targetValues,
|
---|
235 | double originalValue,
|
---|
236 | ReplacementMethodEnum replacementMethod) {
|
---|
237 | OnlineCalculatorError error;
|
---|
238 | var newEstimates = EvaluateModelWithReplacedVariable(model, variableName, modifiableDataset, rows, replacementMethod);
|
---|
239 | var newValue = CalculateValue(targetValues, newEstimates, out error);
|
---|
240 | if (error != OnlineCalculatorError.None) { throw new InvalidOperationException("Error during calculation with replaced inputs."); }
|
---|
241 | return originalValue - newValue;
|
---|
242 | }
|
---|
243 |
|
---|
244 | private static double CalculateImpactForString(string variableName,
|
---|
245 | IRegressionModel model,
|
---|
246 | IDataset problemData,
|
---|
247 | ModifiableDataset modifiableDataset,
|
---|
248 | IEnumerable<int> rows,
|
---|
249 | IEnumerable<double> targetValues,
|
---|
250 | double originalValue,
|
---|
251 | FactorReplacementMethodEnum factorReplacementMethod) {
|
---|
252 |
|
---|
253 | OnlineCalculatorError error;
|
---|
254 | if (factorReplacementMethod == FactorReplacementMethodEnum.Best) {
|
---|
255 | // try replacing with all possible values and find the best replacement value
|
---|
256 | var smallestImpact = double.PositiveInfinity;
|
---|
257 | foreach (var repl in problemData.GetStringValues(variableName, rows).Distinct()) {
|
---|
258 | var originalValues = modifiableDataset.GetReadOnlyStringValues(variableName).ToList();
|
---|
259 | var newEstimates = EvaluateModelWithReplacedVariable(originalValues, model, variableName, modifiableDataset, rows, Enumerable.Repeat(repl, problemData.Rows).ToList());
|
---|
260 | var newValue = CalculateValue(targetValues, newEstimates, out error);
|
---|
261 | if (error != OnlineCalculatorError.None) throw new InvalidOperationException("Error during calculation with replaced inputs.");
|
---|
262 |
|
---|
263 | var curImpact = originalValue - newValue;
|
---|
264 | if (curImpact < smallestImpact) smallestImpact = curImpact;
|
---|
265 | }
|
---|
266 | return smallestImpact;
|
---|
267 | } else {
|
---|
268 | // for replacement methods shuffle and mode
|
---|
269 | // calculate impacts for factor variables
|
---|
270 | var newEstimates = EvaluateModelWithReplacedVariable(model, variableName, modifiableDataset, rows, factorReplacementMethod);
|
---|
271 | var newValue = CalculateValue(targetValues, newEstimates, out error);
|
---|
272 | if (error != OnlineCalculatorError.None) throw new InvalidOperationException("Error during calculation with replaced inputs.");
|
---|
273 |
|
---|
274 | return originalValue - newValue;
|
---|
275 | }
|
---|
276 | }
|
---|
277 |
|
---|
278 | private static IEnumerable<double> EvaluateModelWithReplacedVariable(IRegressionModel model, string variable, ModifiableDataset dataset, IEnumerable<int> rows, ReplacementMethodEnum replacement = ReplacementMethodEnum.Median) {
|
---|
279 | var originalValues = dataset.GetReadOnlyDoubleValues(variable).ToList();
|
---|
280 | double replacementValue;
|
---|
281 | List<double> replacementValues;
|
---|
282 | IRandom rand;
|
---|
283 |
|
---|
284 | switch (replacement) {
|
---|
285 | case ReplacementMethodEnum.Median:
|
---|
286 | replacementValue = rows.Select(r => originalValues[r]).Median();
|
---|
287 | replacementValues = Enumerable.Repeat(replacementValue, dataset.Rows).ToList();
|
---|
288 | break;
|
---|
289 | case ReplacementMethodEnum.Average:
|
---|
290 | replacementValue = rows.Select(r => originalValues[r]).Average();
|
---|
291 | replacementValues = Enumerable.Repeat(replacementValue, dataset.Rows).ToList();
|
---|
292 | break;
|
---|
293 | case ReplacementMethodEnum.Shuffle:
|
---|
294 | // new var has same empirical distribution but the relation to y is broken
|
---|
295 | rand = new FastRandom(31415);
|
---|
296 | // prepare a complete column for the dataset
|
---|
297 | replacementValues = Enumerable.Repeat(double.NaN, dataset.Rows).ToList();
|
---|
298 | // shuffle only the selected rows
|
---|
299 | var shuffledValues = rows.Select(r => originalValues[r]).Shuffle(rand).ToList();
|
---|
300 | int i = 0;
|
---|
301 | // update column values
|
---|
302 | foreach (var r in rows) {
|
---|
303 | replacementValues[r] = shuffledValues[i++];
|
---|
304 | }
|
---|
305 | break;
|
---|
306 | case ReplacementMethodEnum.Noise:
|
---|
307 | var avg = rows.Select(r => originalValues[r]).Average();
|
---|
308 | var stdDev = rows.Select(r => originalValues[r]).StandardDeviation();
|
---|
309 | rand = new FastRandom(31415);
|
---|
310 | // prepare a complete column for the dataset
|
---|
311 | replacementValues = Enumerable.Repeat(double.NaN, dataset.Rows).ToList();
|
---|
312 | // update column values
|
---|
313 | foreach (var r in rows) {
|
---|
314 | replacementValues[r] = NormalDistributedRandom.NextDouble(rand, avg, stdDev);
|
---|
315 | }
|
---|
316 | break;
|
---|
317 |
|
---|
318 | default:
|
---|
319 | throw new ArgumentException(string.Format("ReplacementMethod {0} cannot be handled.", replacement));
|
---|
320 | }
|
---|
321 |
|
---|
322 | return EvaluateModelWithReplacedVariable(originalValues, model, variable, dataset, rows, replacementValues);
|
---|
323 | }
|
---|
324 |
|
---|
325 | private static IEnumerable<double> EvaluateModelWithReplacedVariable(
|
---|
326 | IRegressionModel model, string variable, ModifiableDataset dataset,
|
---|
327 | IEnumerable<int> rows,
|
---|
328 | FactorReplacementMethodEnum replacement = FactorReplacementMethodEnum.Best) {
|
---|
329 | var originalValues = dataset.GetReadOnlyStringValues(variable).ToList();
|
---|
330 | List<string> replacementValues;
|
---|
331 | IRandom rand;
|
---|
332 |
|
---|
333 | switch (replacement) {
|
---|
334 | case FactorReplacementMethodEnum.Mode:
|
---|
335 | var mostCommonValue = rows.Select(r => originalValues[r])
|
---|
336 | .GroupBy(v => v)
|
---|
337 | .OrderByDescending(g => g.Count())
|
---|
338 | .First().Key;
|
---|
339 | replacementValues = Enumerable.Repeat(mostCommonValue, dataset.Rows).ToList();
|
---|
340 | break;
|
---|
341 | case FactorReplacementMethodEnum.Shuffle:
|
---|
342 | // new var has same empirical distribution but the relation to y is broken
|
---|
343 | rand = new FastRandom(31415);
|
---|
344 | // prepare a complete column for the dataset
|
---|
345 | replacementValues = Enumerable.Repeat(string.Empty, dataset.Rows).ToList();
|
---|
346 | // shuffle only the selected rows
|
---|
347 | var shuffledValues = rows.Select(r => originalValues[r]).Shuffle(rand).ToList();
|
---|
348 | int i = 0;
|
---|
349 | // update column values
|
---|
350 | foreach (var r in rows) {
|
---|
351 | replacementValues[r] = shuffledValues[i++];
|
---|
352 | }
|
---|
353 | break;
|
---|
354 | default:
|
---|
355 | throw new ArgumentException(string.Format("FactorReplacementMethod {0} cannot be handled.", replacement));
|
---|
356 | }
|
---|
357 |
|
---|
358 | return EvaluateModelWithReplacedVariable(originalValues, model, variable, dataset, rows, replacementValues);
|
---|
359 | }
|
---|
360 |
|
---|
361 | private static IEnumerable<double> EvaluateModelWithReplacedVariable(IList originalValues, IRegressionModel model, string variable,
|
---|
362 | ModifiableDataset dataset, IEnumerable<int> rows, IList replacementValues) {
|
---|
363 | dataset.ReplaceVariable(variable, replacementValues);
|
---|
364 | //mkommend: ToList is used on purpose to avoid lazy evaluation that could result in wrong estimates due to variable replacements
|
---|
365 | var estimates = model.GetEstimatedValues(dataset, rows).ToList();
|
---|
366 | dataset.ReplaceVariable(variable, originalValues);
|
---|
367 |
|
---|
368 | return estimates;
|
---|
369 | }
|
---|
370 |
|
---|
371 | private static double CalculateValue(IEnumerable<double> targets, IEnumerable<double> estimates, out OnlineCalculatorError error) {
|
---|
372 | calculator.Reset();
|
---|
373 |
|
---|
374 | var targetsEnumerator = targets.GetEnumerator();
|
---|
375 | var estimatesEnumerator = estimates.GetEnumerator();
|
---|
376 |
|
---|
377 | bool targetsHasNextValue = targetsEnumerator.MoveNext();
|
---|
378 | bool estimatesHasNextValue = estimatesEnumerator.MoveNext();
|
---|
379 |
|
---|
380 | while (targetsHasNextValue && estimatesHasNextValue) {
|
---|
381 | calculator.Add(targetsEnumerator.Current, estimatesEnumerator.Current);
|
---|
382 | targetsHasNextValue = targetsEnumerator.MoveNext();
|
---|
383 | estimatesHasNextValue = estimatesEnumerator.MoveNext();
|
---|
384 | }
|
---|
385 |
|
---|
386 | //Check if there is an equal quantity of targets and estimates
|
---|
387 | if (targetsHasNextValue != estimatesHasNextValue) {
|
---|
388 | throw new ArgumentException(string.Format("Targets and Estimates must be of equal length ({0},{1})", targets.Count(), estimates.Count()));
|
---|
389 |
|
---|
390 | }
|
---|
391 |
|
---|
392 | error = calculator.ErrorState;
|
---|
393 | return calculator.Value;
|
---|
394 | }
|
---|
395 | }
|
---|
396 | }
|
---|