-
Notifications
You must be signed in to change notification settings - Fork 0
/
trainLogisticRegression.m
192 lines (174 loc) · 9.22 KB
/
trainLogisticRegression.m
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
function [trainedClassifier, validationAccuracy] = trainLogisticRegression(trainingData, responseData)
% [trainedClassifier, validationAccuracy] = trainClassifier(trainingData,
% responseData)
% Returns a trained classifier and its accuracy. This code recreates the
% classification model trained in Classification Learner app. Use the
% generated code to automate training the same model with new data, or to
% learn how to programmatically train models.
%
% Input:
% trainingData: A matrix with the same number of columns and data type
% as the matrix imported into the app.
%
% responseData: A vector with the same data type as the vector
% imported into the app. The length of responseData and the number of
% rows of trainingData must be equal.
%
% Output:
% trainedClassifier: A struct containing the trained classifier. The
% struct contains various fields with information about the trained
% classifier.
%
% trainedClassifier.predictFcn: A function to make predictions on new
% data.
%
% validationAccuracy: A double containing the accuracy as a
% percentage. In the app, the Models pane displays this overall
% accuracy score for each model.
%
% Use the code to train the model with new data. To retrain your
% classifier, call the function from the command line with your original
% data or new data as the input arguments trainingData and responseData.
%
% For example, to retrain a classifier trained with the original data set T
% and response Y, enter:
% [trainedClassifier, validationAccuracy] = trainClassifier(T, Y)
%
% To make predictions with the returned 'trainedClassifier' on new data T2,
% use
% yfit = trainedClassifier.predictFcn(T2)
%
% T2 must be a matrix containing only the predictor columns used for
% training. For details, enter:
% trainedClassifier.HowToPredict
% Auto-generated by MATLAB on 11-Mar-2022 17:32:29
% Extract predictors and response
% This code processes the data into the right shape for training the
% model.
% Convert input to table
inputTable = array2table(trainingData, 'VariableNames', {'column_1', 'column_2', 'column_3', 'column_4', 'column_5', 'column_6', 'column_7', 'column_8', 'column_9', 'column_10'});
predictorNames = {'column_1', 'column_2', 'column_3', 'column_4', 'column_5', 'column_6', 'column_7', 'column_8', 'column_9', 'column_10'};
predictors = inputTable(:, predictorNames);
response = responseData;
isCategoricalPredictor = [false, false, false, false, false, false, false, false, false, false];
% Train a classifier
% This code specifies all the classifier options and trains the classifier.
% For logistic regression, the response values must be converted to zeros
% and ones because the responses are assumed to follow a binomial
% distribution.
% 1 or true = 'successful' class
% 0 or false = 'failure' class
% NaN - missing response.
successClass = 'nok';
failureClass = 'ok';
% Compute the majority response class. If there is a NaN-prediction from
% fitglm, convert NaN to this majority class label.
numSuccess = sum(response == successClass);
numFailure = sum(response == failureClass);
if numSuccess > numFailure
missingClass = successClass;
else
missingClass = failureClass;
end
responseCategories = {successClass, failureClass};
successFailureAndMissingClasses = categorical({successClass; failureClass; missingClass}, responseCategories);
isMissing = isundefined(response);
zeroOneResponse = double(ismember(response, successClass));
zeroOneResponse(isMissing) = NaN;
% Prepare input arguments to fitglm.
concatenatedPredictorsAndResponse = [predictors, table(zeroOneResponse)];
% Train using fitglm.
GeneralizedLinearModel = fitglm(...
concatenatedPredictorsAndResponse, ...
'Distribution', 'binomial', ...
'link', 'logit');
% Convert predicted probabilities to predicted class labels and scores.
convertSuccessProbsToPredictions = @(p) successFailureAndMissingClasses( ~isnan(p).*( (p<0.5) + 1 ) + isnan(p)*3 );
returnMultipleValuesFcn = @(varargin) varargin{1:max(1,nargout)};
scoresFcn = @(p) [p, 1-p];
predictionsAndScoresFcn = @(p) returnMultipleValuesFcn( convertSuccessProbsToPredictions(p), scoresFcn(p) );
% Create the result struct with predict function
predictorExtractionFcn = @(x) array2table(x, 'VariableNames', predictorNames);
logisticRegressionPredictFcn = @(x) predictionsAndScoresFcn( predict(GeneralizedLinearModel, x) );
trainedClassifier.predictFcn = @(x) logisticRegressionPredictFcn(predictorExtractionFcn(x));
% Add additional fields to the result struct
trainedClassifier.GeneralizedLinearModel = GeneralizedLinearModel;
trainedClassifier.SuccessClass = successClass;
trainedClassifier.FailureClass = failureClass;
trainedClassifier.MissingClass = missingClass;
trainedClassifier.ClassNames = {successClass; failureClass};
trainedClassifier.About = 'This struct is a trained model exported from Classification Learner R2021b.';
trainedClassifier.HowToPredict = sprintf('To make predictions on a new predictor column matrix, X, use: \n yfit = c.predictFcn(X) \nreplacing ''c'' with the name of the variable that is this struct, e.g. ''trainedModel''. \n \nX must contain exactly 10 columns because this model was trained using 10 predictors. \nX must contain only predictor columns in exactly the same order and format as your training \ndata. Do not include the response column or any columns you did not import into the app. \n \nFor more information, see <a href="matlab:helpview(fullfile(docroot, ''stats'', ''stats.map''), ''appclassification_exportmodeltoworkspace'')">How to predict using an exported model</a>.');
% Extract predictors and response
% This code processes the data into the right shape for training the
% model.
% Convert input to table
inputTable = array2table(trainingData, 'VariableNames', {'column_1', 'column_2', 'column_3', 'column_4', 'column_5', 'column_6', 'column_7', 'column_8', 'column_9', 'column_10'});
predictorNames = {'column_1', 'column_2', 'column_3', 'column_4', 'column_5', 'column_6', 'column_7', 'column_8', 'column_9', 'column_10'};
predictors = inputTable(:, predictorNames);
response = responseData;
isCategoricalPredictor = [false, false, false, false, false, false, false, false, false, false];
% Perform cross-validation
KFolds = 5;
cvp = cvpartition(response, 'KFold', KFolds);
% Initialize the predictions to the proper sizes
validationPredictions = response;
numObservations = size(predictors, 1);
numClasses = 2;
validationScores = NaN(numObservations, numClasses);
for fold = 1:KFolds
trainingPredictors = predictors(cvp.training(fold), :);
trainingResponse = response(cvp.training(fold), :);
foldIsCategoricalPredictor = isCategoricalPredictor;
% Train a classifier
% This code specifies all the classifier options and trains the classifier.
% For logistic regression, the response values must be converted to zeros
% and ones because the responses are assumed to follow a binomial
% distribution.
% 1 or true = 'successful' class
% 0 or false = 'failure' class
% NaN - missing response.
successClass = 'nok';
failureClass = 'ok';
% Compute the majority response class. If there is a NaN-prediction from
% fitglm, convert NaN to this majority class label.
numSuccess = sum(trainingResponse == successClass);
numFailure = sum(trainingResponse == failureClass);
if numSuccess > numFailure
missingClass = successClass;
else
missingClass = failureClass;
end
responseCategories = {successClass, failureClass};
successFailureAndMissingClasses = categorical({successClass; failureClass; missingClass}, responseCategories);
isMissing = isundefined(trainingResponse);
zeroOneResponse = double(ismember(trainingResponse, successClass));
zeroOneResponse(isMissing) = NaN;
% Prepare input arguments to fitglm.
concatenatedPredictorsAndResponse = [trainingPredictors, table(zeroOneResponse)];
% Train using fitglm.
GeneralizedLinearModel = fitglm(...
concatenatedPredictorsAndResponse, ...
'Distribution', 'binomial', ...
'link', 'logit');
% Convert predicted probabilities to predicted class labels and scores.
convertSuccessProbsToPredictions = @(p) successFailureAndMissingClasses( ~isnan(p).*( (p<0.5) + 1 ) + isnan(p)*3 );
returnMultipleValuesFcn = @(varargin) varargin{1:max(1,nargout)};
scoresFcn = @(p) [p, 1-p];
predictionsAndScoresFcn = @(p) returnMultipleValuesFcn( convertSuccessProbsToPredictions(p), scoresFcn(p) );
% Create the result struct with predict function
logisticRegressionPredictFcn = @(x) predictionsAndScoresFcn( predict(GeneralizedLinearModel, x) );
validationPredictFcn = @(x) logisticRegressionPredictFcn(x);
% Add additional fields to the result struct
% Compute validation predictions
validationPredictors = predictors(cvp.test(fold), :);
[foldPredictions, foldScores] = validationPredictFcn(validationPredictors);
% Store predictions in the original order
validationPredictions(cvp.test(fold), :) = foldPredictions;
validationScores(cvp.test(fold), :) = foldScores;
end
% Compute validation accuracy
correctPredictions = (validationPredictions == response);
isMissing = ismissing(response);
correctPredictions = correctPredictions(~isMissing);
validationAccuracy = sum(correctPredictions)/length(correctPredictions);