001/* 002 * Copyright (c) 2015-2020, Oracle and/or its affiliates. All rights reserved. 003 * 004 * Licensed under the Apache License, Version 2.0 (the "License"); 005 * you may not use this file except in compliance with the License. 006 * You may obtain a copy of the License at 007 * 008 * http://www.apache.org/licenses/LICENSE-2.0 009 * 010 * Unless required by applicable law or agreed to in writing, software 011 * distributed under the License is distributed on an "AS IS" BASIS, 012 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express implied. 013 * See the License for the specific language governing permissions and 014 * limitations under the License. 015 */ 016 017package org.tribuo.classification.sequence; 018 019import org.tribuo.Prediction; 020import org.tribuo.classification.Label; 021import org.tribuo.classification.evaluation.LabelMetric; 022import org.tribuo.classification.evaluation.LabelMetrics; 023import org.tribuo.evaluation.metrics.MetricID; 024import org.tribuo.evaluation.metrics.MetricTarget; 025import org.tribuo.provenance.EvaluationProvenance; 026import org.tribuo.sequence.AbstractSequenceEvaluator; 027import org.tribuo.sequence.SequenceModel; 028 029import java.util.ArrayList; 030import java.util.HashSet; 031import java.util.List; 032import java.util.Map; 033import java.util.Set; 034 035/** 036 * A sequence evaluator for labels. 037 */ 038public class LabelSequenceEvaluator extends AbstractSequenceEvaluator<Label, LabelMetric.Context, LabelSequenceEvaluation, LabelMetric> { 039 040 @Override 041 protected Set<LabelMetric> createMetrics(SequenceModel<Label> model) { 042 Set<LabelMetric> metrics = new HashSet<>(); 043 // 044 // Populate labelwise values 045 for (Label label : model.getOutputIDInfo().getDomain()) { 046 MetricTarget<Label> tgt = new MetricTarget<>(label); 047 metrics.add(LabelMetrics.TP.forTarget(tgt)); 048 metrics.add(LabelMetrics.FP.forTarget(tgt)); 049 metrics.add(LabelMetrics.TN.forTarget(tgt)); 050 metrics.add(LabelMetrics.FN.forTarget(tgt)); 051 metrics.add(LabelMetrics.PRECISION.forTarget(tgt)); 052 metrics.add(LabelMetrics.RECALL.forTarget(tgt)); 053 metrics.add(LabelMetrics.F1.forTarget(tgt)); 054 metrics.add(LabelMetrics.ACCURACY.forTarget(tgt)); 055 } 056 057 // 058 // Populate averaged values. 059 MetricTarget<Label> micro = MetricTarget.microAverageTarget(); 060 metrics.add(LabelMetrics.TP.forTarget(micro)); 061 metrics.add(LabelMetrics.FP.forTarget(micro)); 062 metrics.add(LabelMetrics.TN.forTarget(micro)); 063 metrics.add(LabelMetrics.FN.forTarget(micro)); 064 metrics.add(LabelMetrics.PRECISION.forTarget(micro)); 065 metrics.add(LabelMetrics.RECALL.forTarget(micro)); 066 metrics.add(LabelMetrics.F1.forTarget(micro)); 067 metrics.add(LabelMetrics.ACCURACY.forTarget(micro)); 068 069 MetricTarget<Label> macro = MetricTarget.macroAverageTarget(); 070 metrics.add(LabelMetrics.TP.forTarget(macro)); 071 metrics.add(LabelMetrics.FP.forTarget(macro)); 072 metrics.add(LabelMetrics.TN.forTarget(macro)); 073 metrics.add(LabelMetrics.FN.forTarget(macro)); 074 metrics.add(LabelMetrics.PRECISION.forTarget(macro)); 075 metrics.add(LabelMetrics.RECALL.forTarget(macro)); 076 metrics.add(LabelMetrics.F1.forTarget(macro)); 077 metrics.add(LabelMetrics.ACCURACY.forTarget(macro)); 078 079 // Target doesn't matter for balanced error rate, so we just use 080 // average.macro as it's the macro average of recalls. 081 metrics.add(LabelMetrics.BALANCED_ERROR_RATE.forTarget(macro)); 082 083 return metrics; 084 } 085 086 @Override 087 protected LabelMetric.Context createContext(SequenceModel<Label> model, List<List<Prediction<Label>>> predictions) { 088 // Warning this passes a null in as the model. 089 return new LabelMetric.Context(model, flattenList(predictions)); 090 } 091 092 @Override 093 protected LabelSequenceEvaluation createEvaluation(LabelMetric.Context ctx, 094 Map<MetricID<Label>, Double> results, 095 EvaluationProvenance provenance) { 096 return new LabelSequenceEvaluation(results, ctx, provenance); 097 } 098 099 private static List<Prediction<Label>> flattenList(List<List<Prediction<Label>>> predictions) { 100 List<Prediction<Label>> flatList = new ArrayList<>(); 101 102 for (List<Prediction<Label>> list : predictions) { 103 flatList.addAll(list); 104 } 105 106 return flatList; 107 } 108}