TIME SERIES DATA WAVEFORM ANALYZER, METHOD, AND PROGRAM
PROBLEM TO BE SOLVED: To provide a time series data waveform analyzer in which a calculation amount is reduced, and a method and a program.SOLUTION: A time series data waveform analyzer related to one embodiment, comprises a feature waveform addition section, a feature quantity calculation section,...
Saved in:
Main Authors | , |
---|---|
Format | Patent |
Language | English Japanese |
Published |
10.08.2017
|
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | PROBLEM TO BE SOLVED: To provide a time series data waveform analyzer in which a calculation amount is reduced, and a method and a program.SOLUTION: A time series data waveform analyzer related to one embodiment, comprises a feature waveform addition section, a feature quantity calculation section, a parameter update section, and a feature waveform removal section. The feature waveform addition section adds a part of partial time series included in time series data with labels as a feature waveform to a feature waveform aggregate. The feature quantity calculation section extracts the time series data with one or a plurality of labels at random, and calculates feature quantity of the feature waveform with respect to the time series data with the labels extracted on the basis of a TSS method. The parameter update section updates a parameter including the feature waveform and a weighting factor corresponding to the feature waveform based on a stochastic gradient descent method, on the basis of the feature quantity. The feature waveform removal section removes the feature waveform having a corresponding weighting factor of 0 from the feature waveform aggregate. The evaluation function generation section generates an evaluation function based on the feature waveform and the weighting factor.SELECTED DRAWING: Figure 1
【課題】計算量が低減された時系列データ波形分析装置、方法、及びプログラムを提供する。【解決手段】一実施形態に係る時系列データ波形分析装置は、特徴波形追加部と、特徴量計算部と、パラメータ更新部と、特徴波形除去部と、を備える。特徴波形追加部は、ラベル付き時系列データに含まれる部分時系列の一部を、特徴波形集合に特徴波形として追加する。特徴量計算部は、1つ又は複数のラベル付き時系列データをランダムに抽出し、抽出されたラベル付き時系列データに対する、特徴波形の特徴量をTSS法に基づいて計算する。パラメータ更新部は、特徴量に基づいて、特徴波形と、特徴波形に対応する重み係数と、を含むパラメータを確率的勾配降下法に基づいて更新する。特徴波形除去部は、対応する重み係数が0である特徴波形を、特徴波形集合から除去する。評価関数生成部は、特徴波形と、重み係数と、に基づいて、評価関数を生成する。【選択図】図1 |
---|---|
Bibliography: | Application Number: JP20160021313 |