Information processing device, information processing method, and program

Abstract

To improve the accuracy of estimation using an estimation model for time series data.SOLUTION: An information processing device includes a processing unit. The processing unit adjusts the first distance between first time-series data and a plurality of identification patterns corresponding to a plurality of classes by using the degree of presence that indicates the degree to which a plurality of points in time in the first time-series data is points in time at which one of the plurality of identification patterns is present. The processing unit calculates a first feature quantity representing a feature of the first time-series data by using the adjusted first distance. The processing unit inputs a first feature quantity to an estimation model in which parameters are learned together with similar identification patterns and the degree of presence so as to reduce estimation errors, and thereby obtains identification patterns, classes, and the degree of presence similar to the first time-series data.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] An embodiment of the present invention relates to an information processing device, an information processing method, and a program. [Background technology]

[0002] In classification technology for time-series data (time-series waveform data) such as sensor data, it is desirable to clarify the basis for classification in addition to improving classification performance (estimation accuracy).As a classification technology for time-series data that clarifies the basis for classification, a supervised shapelet learning method has been proposed that learns one or more shapelets, which are short waveform patterns (discrimination patterns) effective for classification, together with a classifier (estimation model), and has attracted attention in recent years in the fields of data mining and machine learning. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] International Publication No. 2020 / 179188 [Non-patent literature]

[0004] [Non-Patent Document 1] Nestor Cabello et al., “Fast and Accurate Time Series Classification Through Supervised Interval Search”, ICDM 2020 : IEEE International Conference on Data Mining, Pages:948-953. [Non-Patent Document 2] Xuan-May Thi Le et al., “Learning Perceptual Position-Aware Shapelets for Time Series Classification”, ECML PKDD 2020 : The European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases. Summary of the Invention [Problem to be solved by the invention]

[0005] An object of the present invention is to provide an information processing device, an information processing method, and a program that can improve the accuracy of estimation using an estimation model for time-series data. [Means for solving the problem]

[0006] An information processing device according to an embodiment includes a processor. The processor adjusts a first distance between first time series data and a plurality of discrimination patterns corresponding to a plurality of classes, using an existence degree that indicates a degree to which a plurality of time points in the first time series data are times at which any of the plurality of discrimination patterns exists. The processor calculates a first feature amount that indicates a feature of the first time series data, using the adjusted first distance. The processor obtains a discrimination pattern similar to the first time series data, a class, and an existence degree by inputting the first feature amount into an estimation model in which parameters are learned together with similar discrimination patterns and existence degrees so as to reduce an estimation error. [Brief description of the drawings]

[0007] [Figure 1] FIG. 1 is a block diagram of an information processing apparatus according to a first embodiment. [Diagram 2] 5 is a flowchart of a learning process according to the first embodiment. [Diagram 3] 4 is a flowchart of an estimation process according to the first embodiment. [Figure 4]FIG. 2 is a diagram showing an example of a display screen according to the first embodiment. [Diagram 5] FIG. 13 is a diagram showing an example of a display screen according to a comparative example. [Figure 6] FIG. 11 is a block diagram of an information processing apparatus according to a second embodiment. [Figure 7] 10 is a flowchart of a learning process according to the second embodiment. [Figure 8] 10 is a flowchart of an estimation process according to a second embodiment. [Figure 9] FIG. 11 is a diagram showing an example of a display screen according to the second embodiment. [Figure 10] FIG. 2 is a hardware configuration diagram of an information processing apparatus according to the first or second embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0008] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, preferred embodiments of an information processing apparatus according to the present invention will be described in detail with reference to the accompanying drawings.

[0009] The following embodiment can be applied to a system that analyzes time-series data such as sensor data detected and collected by various sensors, but is not limited to the following systems. A system for classifying time series data into one of a plurality of classes. The plurality of classes includes, for example, a normal class indicating that the time series data is normal and an abnormal class indicating that the time series data is abnormal.

[0010] In conventional shapelet learning methods, the distance (similarity) between each shapelet and the most compatible (similar) partial time series data (partial waveform) in the time series data is used as a feature. In this case, the existence region of the shapelet is basically the entire time series data. The existence region can be interpreted as the region of the time series data where the shapelet may exist and is the region to be matched with the shapelet.

[0011] The shapelet learning method can discover shapelets that allow for positional deviation. However, no technology has yet been proposed to identify in which area of ​​time-series data a shapelet appears (exists). If a search is performed for all points (times) on the time-series data to see if the point is in an area of ​​existence, then if the length of the time-series data (the number of points included in the time-series data) is Q, then it is necessary to learn the shapelet 2 to the power of Q times, which is practically impossible in terms of computation time.

[0012] Although a technique for giving the existence region of a shapelet in advance has been proposed, such a technique cannot discover the existence region itself.

[0013] In the following embodiment, a classifier (an example of an estimation model) trained to identify multiple existence regions for each shapelet is used to estimate the existence region along with the classification result of time-series data. For example, when the existence region of a shapelet is important for class classification, the classification performance can be improved. In addition, by displaying (visualizing, presenting) the existence region together with the shapelet that is the basis for classification, more descriptive information can be provided to the user.

[0014] For example, the embodiment has the following features. A function to adjust the distance between the matching shapelet and partial time series data using the presence degree that indicates the degree to which the shapelet exists in the area. A function to update the presence of shapelets so that the presence area of ​​shapelets can be visually understood while reducing the classification loss by the classifier (reducing the estimation error by the estimation model).

[0015] In the following, an example will be described in which shapelets are used as discrimination patterns for classification, but the discrimination patterns may be patterns other than shapelets.

[0016] Here, the terms and symbols used in the following description will be explained.

[0017] Hereinafter, let the time series dataset be T, the number of time series data included in the time series dataset T be I, the length of each time series data be Q points, and the number of shapelets be K. The length of each shapelet is given as L points (where L satisfies L < Q). For simplicity of explanation here, it is assumed that the length of each time series data is the same Q points, and the length of each shapelet is the same L points. When the length of each time series data is the same Q points, the time series dataset T can be considered as an I×Q continuous value matrix. The extension to the case where the lengths are different is easy for each case.

[0018] Also, hereinafter, the case where the time series data is univariate time series data will be described as an example. In the case of multivariate time series data, for example, the method described below can be applied for each variable, and it can be easily extended to identify shapelets and the existence regions of shapelets.

[0019] When the length of each shapelet is the same L points, the K shapelets (hereinafter referred to as shapelet S) are a K×L continuous value matrix. The existence regions of the K shapelets are represented by the existence degree M, which is a K×J continuous value matrix. J = Q - L + 1, which represents the length of the region in the time series data length (Q) where the shapelets can be searched. Hereinafter, the existence degree of the k-th shapelet (where k is an integer satisfying 1 ≤ k ≤ K) (hereinafter referred to as shapelet S k is denoted as m k and is a J-dimensional vector. Also, hereinafter, the existence degree (scalar value) of the j-th point (where j is an integer satisfying 1 ≤ j ≤ J) of the k-th shapelet is denoted as m k k,j and is denoted as m

[0020] The existence degree m k,j represents that, for example, the smaller the value, the lower the degree of the shapelet being in the existence region. The existence degree may also represent that the larger the value, the lower the degree of the shapelet being in the existence region. Hereinafter, the existence degree that represents that the larger the value, the lower the degree of the shapelet being in the existence region is defined as the penalty degree m'.​k,j The penalty level m' k,j For example, the abundance m k,j The ReLu function in equation (1) is expressed by the following equation (2).

number

number

[0021] In formula (1), the penalty level is calculated to take a value in the range of [1,∞]. k,j is the shapelet S at the j-th point of the time series data. k It corresponds to the value of the penalty level (scalar value) when exists. k Penalty level m' k is the J-th penalty level m' k,j It becomes a J-th vector with elements.

[0022] Penalty degree m' k,j If the value of is small, the jth point of the time series data is k This is interpreted as a high degree of existence of the area of ​​the penalty m'. k,j If the value of is large, the jth point of the time series data is k This is interpreted as meaning that the degree to which this is an area of ​​existence is small.

[0023] As described below, in an embodiment, a penalty level m′ is used in conjunction with a number of shapelets S. k (abundance m k ) is learned.

[0024] (First embodiment) Fig. 1 is a block diagram showing an example of a configuration of an information processing device 100 according to a first embodiment. As shown in Fig. 1, the information processing device 100 includes a reception unit 101, an output control unit 102, a distance calculation unit 111, a feature calculation unit 112, a learning unit 113, an estimation unit 114, a storage unit 131, and a display unit 132.

[0025] The receiving unit 101 receives input of various information used in the information processing device 100. For example, the receiving unit 101 receives designation of time-series data used for learning by the learning unit 113, time-series data to be estimated, and various parameters used in learning or estimation.

[0026] The parameters are, for example, the following information: · Number of shapelets K Shapelet length L

[0027] The distance calculation unit 111 calculates a distance DA (first distance) between time series data TSA (first time series data) input as an estimation target and a plurality of shapelets S. During learning, the distance calculation unit 111 calculates a distance DB (second distance) between time series data TSB (second time series data) input for learning processing and a plurality of shapelets.

[0028] Here, the i-th (i is an integer satisfying 1≦i≦I) time series data and the k-th shapelet S k The definition of the distance between the shapelet and the offset j is explained below. The distance between the time series data and the shapelet is defined as the minimum value among J distances calculated by changing the offset j in the range from 1 to J and calculating the distance between the shapelet and the partial time series data extracted based on each offset. The offset corresponds to a reference point of the partial time series data included in the time series data that is to be matched with the shapelet. The reference point is, for example, the start point (earliest time) of the partial time series data, but may be any other point such as the center point or the end point.

[0029] In this way, the shapelet is located at a position close to (similar to) the partial time series data included in the time series data. Note that the distance is measured, for example, by Euclidean distance.

[0030] The distance calculation unit 111 calculates the shapelet s kand the i-th time series data t i The partial time series data corresponding to the offset j in and the distance d i,k,j (s k ,t i ) is calculated. Note that L k represents the length of the k-th shapelet, but if we set each shapelet to the same length, L k has the same value L.

number

[0031] The distance is not limited to the distance calculated by the formula (3), and any other type of distance may be used. For example, the distance may be calculated by the DTW (Dynamic Time-Warping) distance instead of the Euclidean distance.

[0032] The distance calculation unit 111 further calculates the penalty level m' k The distance (distances DA, DB) are adjusted using the above. Hereinafter, the adjusted distance may be referred to as an adjusted distance. For example, the distance calculation unit 111 adjusts the distance by a penalty level m' k The adjusted distance is calculated by multiplying the distance by .

[0033] As mentioned above, the penalty level is calculated to take a value in the range of [1, ∞]. Therefore, the penalty level m' k By multiplying by , the distance calculation unit 111 calculates the adjusted distance so that the greater the penalty level, the greater the adjusted distance. As a result, each shapelet becomes more likely to match points on time-series data with a low penalty level, and the matching points on the time-series data become the existence area of ​​the shapelet.

[0034] The feature calculation unit 112 calculates a feature amount representing the feature of the time series data by using the adjusted distance. For example, the feature calculation unit 112 calculates a feature amount FA (first feature amount) representing the feature of the time series data TSA by using the adjusted distance calculated from the time series data TSA. During learning, the feature calculation unit 112 calculates a feature amount FB (second feature amount) representing the feature of the time series data TSB by using the adjusted distance calculated from the time series data TSB.

[0035] Specifically, the feature calculation unit 112 calculates the shapelet s k Using the adjusted distance calculated for each k For each i-th time series data t i Feature quantity x that represents the characteristics of i,k The feature calculation unit 112 calculates the feature amount x i,k Calculate.

number

[0036] In addition, J k is the k-th shapelet s k represents the length of the region to be searched (the range of offsets). If each shapelet is of the same length, J k has the same value J.

[0037] As shown in formula (4), the feature calculation unit 112 calculates the feature amount by aggregating the adjusted distances in the entire offset range of the time series data for each shapelet. Formula (4) shows an example in which the minimum value of multiple adjusted distances corresponding to multiple offsets is used as the feature amount.

[0038] The distance corresponding to equation (3) can also be regarded as dissimilarity. Therefore, similarity may be used instead of distance. For example, similarity is expressed as k and the i-th time series data t imay be defined to take a value of 1 if the partial time series data corresponding to offset j of is sufficiently similar to the partial time series data corresponding to offset j of

[0039] Also, the feature amount may be calculated by taking the sum over the offset j, instead of taking the minimum value over the offset j as in equation (4). In this case, the penalty degree m' k Instead of multiplying by m', the penalty for distance is k As in this example, a feature can be defined using the number of partial time series data similar to a shapelet on the time series data.

[0040] The learning unit 113 executes a learning process of the classifier using the time series data TSB for learning. For example, the learning unit 113 learns the parameters of the classifier, the plurality of shapelets, and the penalty degree m' so as to reduce the classification loss. k A learning process is performed to update the

[0041] For example, the learning unit 113 executes the learning process according to the following formula (5). Formula (5) is a function of multiple time series data t i Each loss G i W is the sum of the classifier parameters w c,k (weights). c represents a set of C classes (C is an integer equal to or greater than 2).

number

[0042] Loss G in Eq. (5) i is expressed by the following equation (6). Also, y^ in equation (6) i,c corresponds to the output of the classifier and is expressed by the following equation (7). Note that y^ in equation (6) i,c_i is the correct class c i This corresponds to the output of the classifier for (c_i). In this manner, the learning unit 113 executes the learning process so as to reduce the error between the class output by the classifier and the correct class.

number

number

[0043] The classifier is not limited to a model that outputs a classification result as shown in equation (7), but may be any model that inputs features to estimate a class and can learn parameters, including shapelets and penalty levels (presence levels), to reduce the estimation error.

[0044] In addition, the loss G in equation (6) i is the abundance m k The second term corresponds to a term (regularization term) for updating the presence so that the values ​​of the presences at adjacent times are close to each other. The second term can be interpreted as a term for updating the presence so that the shapelet's existence region becomes as continuous and wide as possible on the time series data. Note that the loss G i If the second term is not included, the region in which the shapelet exists may not be a region that includes a plurality of continuous points.

[0045] The learning unit 113 may, for example, adjust the shapelet S and the classifier parameters w to reduce the classification loss according to equation (5). c,k (Shapelet classifier update unit). The method for updating the classifier may be any conventional method, but for example, a method of slightly updating the parameter values ​​in a direction that reduces (reduces) the classification loss by using the gradient descent method can be applied (for example, see Non-Patent Document 2).

[0046] Furthermore, the learning unit 113 updates the penalty level at the matching point on the time-series data for each shapelet so as to reduce the classification loss (penalty level update unit). As for the method of updating the penalty level, a method using, for example, gradient descent can be applied, as in the case of the shapelet S and the classifier parameters.

[0047] The estimation unit 114 executes an estimation process for the time series data TSA input as an estimation target, using the classifier trained by the learning unit 113. For example, the estimation unit 114 inputs the feature amount FA calculated for the time series data TSA to the trained classifier, and estimates the class estimated by the classifier, the shapelet corresponding to the class, and the penalty degree m' k And asks.

[0048] The classifier is, for example, y^ calculated by equation (7). i,c The class c with the largest value may be output as the classification result, or the probability of each class c may be output as the classification result. The probability of each class c can be calculated, for example, by the formula written as an argument of the log function of formula (6).

[0049] The output control unit 102 controls the output of various information used in the information processing device 100. For example, the output control unit 102 outputs (displays) information indicating the estimation result by the estimation unit 114 to the display unit 132. The estimation result includes, for example, a shapelet output for the feature amount FA, a class which is a classification result, and a penalty level m' k Including the penalty level m' k corresponds to information indicating whether or not the area is an existence area.

[0050] At least a part of each of the above units (reception unit 101, output control unit 102, distance calculation unit 111, feature calculation unit 112, learning unit 113, and estimation unit 114) may be realized by one or more processing units. Each of the above units is realized, for example, by one or more processors. For example, each of the above units may be realized by having a processor such as a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit) execute a program, that is, by software. Each of the above units may be realized by a processor such as a dedicated IC (Integrated Circuit), that is, by hardware. Each of the above units may be realized by using both software and hardware. When multiple processors are used, each processor may realize one of the units, or two or more of the units.

[0051] The storage unit 131 stores various information used in the information processing device. For example, the storage unit 131 stores time series data TSB used for learning, time series data TSA to be estimated, and various parameters (such as the number K of shapelets and the length L of the shapelets). The storage unit 131 also stores parameters (such as weights) of the classifier obtained by the learning process, the shapelets, and the penalty level.

[0052] The storage unit 131 can be configured from any commonly used storage medium, such as a flash memory, a memory card, a RAM (Random Access Memory), a HDD (Hard Disk Drive), or an optical disk.

[0053] The display unit 132 is an example of a device that displays various information used in the information processing device 100. The display unit 132 is realized by a display device such as a liquid crystal display.

[0054] The information processing device 100 may be physically configured by one device, or may be physically configured by multiple devices. For example, the information processing device 100 may be constructed on a cloud environment. Furthermore, each unit in the information processing device 100 may be distributed and provided in multiple devices. For example, the information processing device 100 (information processing system) may be configured to include a device (e.g., a learning device) having a function required for learning (e.g., the learning unit 113) and a device (e.g., an estimation device) having a function required for estimation (e.g., the estimation unit 114).

[0055] Next, a flow of a learning process by the information processing device 100 of the first embodiment will be described. Fig. 2 is a flowchart showing an example of the learning process in the first embodiment. The purpose of the learning process is to identify one or more shapelets that are waveform patterns (discrimination patterns) effective for classification, the existence region (penalty level) of each shapelet on the time-series data, and the parameters of the classifier.

[0056] The receiving unit 101 receives an input of a time-series data set including time-series data TSB for learning (step S101). The time-series data TSB is data to which a label indicating a correct class is assigned. The receiving unit 101 may receive an input of parameters such as the number K of shapelets and the length L of the shapelets. The receiving unit 101 may be configured not to receive an input of parameters but to use default values. The default value of the number K of shapelets is, for example, 20. The default value of the length L of the shapelet is calculated, for example, by L=Q×(value indicating a proportion). The value indicating the proportion may be set to a fixed value such as 0.1.

[0057] The learning unit 113 initializes the shapelet S, the penalty level, and the parameters of the classifier (step S102). The simplest initialization method is to set values ​​randomly.

[0058] Thereafter, the following steps (steps S103 to S109) are repeated to learn the shapelet S, the penalty level, and the classifier parameters.

[0059] First, the distance calculation unit 111 calculates the distance between each of the I pieces of time-series data and the K pieces of shapelets (step S103). The distance calculation unit 111 calculates an adjusted distance adjusted using the penalty level for each point of the time-series data (step S104).

[0060] The feature calculation unit 112 calculates the feature amount from the adjusted distance using, for example, equation (4) (step S105).

[0061] The learning unit 113 calculates a classification loss from the calculated feature amount and the parameters of the classifier. Then, the learning unit 113 updates the shapelet and the parameters of the classifier so as to reduce the classification loss (step S106). Furthermore, the learning unit 113 updates the penalty level so as to reduce the classification loss (step S107). Furthermore, the learning unit 113 updates the penalty level so that the penalty levels of adjacent points approach the same value (step S108).

[0062] The learning unit 113 determines whether or not to end the repetition (step S109). For example, the learning unit 113 determines to end the repetition when the number of repetitions reaches an upper limit. If it is determined not to end the repetition (step S109: No), the process returns to step S103 and is repeated.

[0063] If it is determined that the repetition is to be ended (step S109: Yes), the learning process ends. Note that the learning unit 113 stores in the storage unit 131 the shapelet S, the penalty level, and the classifier parameters obtained by the learning process.

[0064] Next, a flow of estimation processing by the information processing device 100 of the first embodiment will be described. Fig. 3 is a flowchart showing an example of the estimation processing in the first embodiment. The estimation processing aims to classify the time series data TSA to be estimated using a classifier (estimation model), shapelets, and a penalty level learned by a learning process, and to visualize the classification result, the shapelets that contributed to the classification, and the existence area of ​​the shapelets together with the time series data.

[0065] The reception unit 101 receives input of time series data TSA to be estimated, and information on a trained classifier (estimation model), shapelet, and penalty level (step S201). Note that in the estimation process, unlike in the learning process, initialization of the shapelet S, penalty level, and classifier parameters is not performed, and the input information is used as is.

[0066] Steps S202 to S204 for calculating features for time-series data are similar to steps S103 and S105 of the learning process.

[0067] The estimation unit 114 performs class classification using the calculated feature amount and the classifier parameters (step S205).

[0068] The output control unit 102 outputs the estimation result by the estimation unit 114 (step S206) and ends the estimation process. For example, the output control unit 102 displays on the display unit 132 a display screen that visualizes the shapelets that contributed to the classification and the existence areas of the shapelets that contributed to the classification together with time-series data.

[0069] Fig. 4 is a diagram showing an example of a display screen of the first embodiment. Fig. 4 shows an example of visualization of a case where the number of shapelets 402 that contributed to classification is one, and the existence region 403 of the shapelet 402 is specified as one region near the center of the time-series data 401. The bar graph at the bottom of Fig. 4 shows an example of distribution of feature amounts (distance from the shapelet) when classifying a large number of time-series data.

[0070] The presence region 403 is specified as, for example, a point (time) where the penalty level is equal to or lower than a threshold. If learning is performed using a loss including a term for updating the presence levels (penalty levels) at adjacent times so that they become closer to each other, as in equation (6), the presence region can be made as wide and continuous as possible on the time series data.

[0071] Fig. 5 is a diagram showing an example of a display screen according to a comparative example in which the existence region of a shapelet is not learned. Fig. 5 shows an example in which a shapelet 502 is visualized on time-series data 501 when the number of shapelets 502 that contributed to classification is one. For convenience of explanation, an existence region 503 equivalent to the entire region of the time-series data 501 is displayed, but the existence region 503 does not have to be visualized.

[0072] In the time series data of Figures 4 and 5, the slopes of the upward-pointing waveforms near the center are different, but the slopes of the upward-pointing waveforms near both ends are the same. In other words, Figures 4 and 5 are examples of classification into one of two classes (for example, a normal class or an abnormal class) based on partial time series data near the center.

[0073] In a comparative example (Figure 5) where the existence region is not learned, the shapelet can be adapted to any region on the time series data. Therefore, for example, the shapelet may be adapted incorrectly to time series data with superimposed noise. Figure 5 shows an example where the shapelet is adapted incorrectly to a waveform near the left end, rather than to a waveform near the center. As a result, the shapelet is not learned correctly, and the estimation process cannot sufficiently separate the features between the two classes for the time series data to be estimated. The bar graph at the bottom of Figure 5 shows how the distribution of features is not separated.

[0074] In contrast, in this embodiment (FIG. 4) in which the existence region is learned, the existence region of the shapelet is identified while the shapelet is being learned. In the example of FIG. 4, the existence region excluding the vicinity of both the left and right ends is learned. This allows the shapelet to fit the central waveform even in time series data with superimposed noise. As a result, the shapelet is also correctly learned, and the feature amounts between the two classes can be more appropriately separated even in the time series data to be estimated. In addition, it is possible to present to the user information that the shapelet effective for classification exists near the center of the time series data and does not exist at both ends.

[0075] In this way, in the first embodiment, in addition to learning the shapelets and the classifier, one or more existence regions on the time series data in which each shapelet exists are simultaneously and efficiently learned. This makes it possible to improve the accuracy of estimation using an estimation model for time series data. In addition, by displaying (visualizing, presenting) not only the shapelets that serve as the basis for classification but also the existence regions, it becomes possible to provide the user with more descriptive information.

[0076] Second Embodiment The distance between each shapelet and partial time series data may include multiple types (components) of distance, such as a distance related to the shape of the waveform, a distance related to the height of the waveform, and a distance related to the magnitude of the waveform. In the conventional shapelet learning method, the use of such multiple types of distances and the degree to which the multiple types of distances contribute to classification (hereinafter, the contribution degree) are not taken into consideration.

[0077] The information processing device of the second embodiment classifies time-series data using a plurality of types of distances, and visualizes the contribution of each classification of the plurality of types of distances.

[0078] Fig. 6 is a block diagram showing an example of the configuration of an information processing device 100-2 according to the second embodiment. As shown in Fig. 6, the information processing device 100-2 includes a reception unit 101, an output control unit 102-2, a distance calculation unit 111-2, a feature calculation unit 112-2, a learning unit 113-2, an estimation unit 114, a storage unit 131, and a display unit 132.

[0079] In the second embodiment, the functions of the output control unit 102-2, the distance calculation unit 111-2, the feature calculation unit 112-2, and the learning unit 113-2 are different from those in the first embodiment. The other configurations and functions are the same as those in FIG. 1, which is a block diagram of the information processing device 100 in the first embodiment, so they are denoted by the same reference numerals and will not be described here.

[0080] The distance calculation unit 111-2 calculates the distances (distance DA, distance DB) between the time series data and multiple shapelets. The distance calculation unit 111-2 first calculates multiple types of distances. Hereinafter, each of the multiple types of distances may be referred to as a distance component. The distance calculation unit 111-2 calculates the distance DA and distance DB by integrating multiple distance components using the contribution degree. As a method of integration using the contribution degree, for example, a method of calculating an average value obtained by weighting each distance component by the contribution degree as the distance DA or distance DB can be applied.

[0081] Below are some examples of distance types: The difference between the statistics of the partial time series data and the statistics of the shapelet. The statistics are, for example, the average value, standard deviation, or variance of the values ​​of each point. -The distance between the shape of partial time series data and the shapelet shape. -The distance between the shape of the standardized partial time series data and the shapelet shape. The distance between the frequency components of the partial time series data and the frequency components of the shapelet. The distance between the pth (p is an integer greater than or equal to 1) derivative of the partial time series data and the pth derivative of the shapelet.

[0082] The difference in the average value can be interpreted as the shift in the height direction of the sensor value (the height direction of the waveform). The difference in the standard deviation or the difference in the variance can be interpreted as the difference in the magnitude of the amplitude. The distance between shapes may be Euclidean distance or DTW (Dynamic Time-Warping) distance. Standardization means, for example, eliminating the shift in the height direction of the sensor value (i.e., setting the average value of the partial time series data to 0) and aligning the amplitude (i.e., setting the standard deviation of the partial time series data to 1).

[0083] In this embodiment, of the above-mentioned multiple types of distances (distance components), two or more types of distances are used. In the following, the number of distance components is assumed to be N (N is an integer equal to or greater than 2). k Since N contributions to are parameters, the contribution V is a K×N continuous value matrix. In the following, the contribution of the nth (n is an integer satisfying 1≦n≦N) is defined as v n Contribution v n is a K-dimensional vector.

[0084] For example, the distance calculation unit 111-2 calculates a distance d i,k,j (s k ,t i ) is calculated. Note that formula (8) is an example when N=3.

number

[0085] (8) In Eq. normalized is a standardized partial time series data, and is calculated, for example, by the following equation (9).

number

[0086] (8) In Eq. meam and s stdare the mean and variance of the shapelet, and are calculated, for example, by the following equations (10) and (11), respectively.

number

number

[0087] (8) In Eq. meam and s std are the mean and variance of the partial time series data, and are calculated, for example, by the following equations (12) and (13), respectively.

number

number

[0088] (8) v~ in Eq. n is the normalized contribution v n and is calculated, for example, by the following formula (14).

number

[0089] The feature calculation unit 112-2 calculates the feature quantity x i,k The expression (15) uses the contribution degree V, for example, the distance calculated by the expression (8), which is different from the feature quantity x i,k This differs from equation (4) which calculates

number

[0090] The learning unit 113-2 tunes the parameters of the classifier, the multiple shapelets, and the penalty level m' so as to reduce the classification loss. k A learning process is performed to update the contributions V and V.

[0091] For example, the learning unit 113-2 executes the learning process according to the following equation (16). i is the same as the above formula (6). In this way, the learning unit 113-2 executes a learning process further including updating the contribution degree so as to reduce the error between the class output by the classifier and the correct class. As a method for updating the contribution degree V, a method using, for example, the gradient descent method can be applied, as in the case of the shapelet S, the classifier parameters, and the penalty degree.

number

[0092] The output control unit 102-2 outputs the shapelet, class, and penalty level m' obtained by the estimation process. k In addition, the output control unit 102 differs from the output control unit 102 of the first embodiment in that it further outputs the contribution degree.

[0093] Next, the learning process by the information processing device 100-2 of the second embodiment will be described with reference to Fig. 7. Fig. 7 is a flowchart showing an example of the learning process in the second embodiment. In this embodiment, in addition to the shapelet S, the penalty level, and the classifier parameters, the contribution level V of the distance component is learned.

[0094] The receiving unit 101 receives an input of a time-series data set including time-series data TSB for learning (step S301), similarly to step S101 (FIG. 2) in the first embodiment.

[0095] The learning unit 113-2 initializes the shapelet S, the penalty level, and the classifier parameters, as well as the contribution level (step S302). The simplest initialization method is to set values ​​randomly.

[0096] Thereafter, the following steps (steps S303 to S311) are repeated to learn the shapelet S, the penalty level, and the parameters and contribution levels of the classifier.

[0097] The distance calculation unit 111-2 calculates multiple types of distances between each of the I time series data and the K shapelets (step S303). The distance calculation unit 111-2 integrates the multiple types of distances using the contribution degree (step S304). The distance calculation unit 111-2 calculates a distance in which the multiple types of distances (distance components) are weighted by the contribution degree, for example, using equation (8).

[0098] Steps S305 to S308 are similar to steps S104 to S107 in the information processing device 100 of the first embodiment, and therefore a description thereof will be omitted.

[0099] In this embodiment, the learning unit 113-2 updates the contribution of each distance component so as to reduce the classification loss (step S309).

[0100] Steps S310 to S311 are the same as steps S108 to S109 in the information processing device 100 of the first embodiment, and therefore a description thereof will be omitted.

[0101] The learning unit 113-2 stores in the storage unit 131 the shapelet S, the penalty level, the classifier parameters, and the contribution level obtained by the learning process.

[0102] Next, the estimation process by the information processing device 100-2 of the second embodiment will be described with reference to Fig. 8. Fig. 8 is a flowchart showing an example of the estimation process in the second embodiment. In this embodiment, in addition to the classification result, the shapelets that contributed to the classification, and the existence area of ​​the shapelets, the contribution degree is visualized.

[0103] The receiving unit 101 receives input of the time-series data TSA to be estimated, and information on a trained classifier (estimation model), shapelets, and a penalty level, similarly to step S201 (FIG. 3) in the first embodiment (step S401).

[0104] Steps S402 to S405 for calculating feature amounts for time-series data are similar to steps S303 and S306 in the learning process of FIG.

[0105] The estimation unit 114 performs class classification using the calculated feature amount and the classifier parameters (step S406).

[0106] The output control unit 102-2 outputs the estimation result by the estimation unit 114 (step S407) and ends the estimation process. For example, the output control unit 102-2 displays on the display unit 132 a display screen that visualizes the shapelets that contributed to the classification, the existence regions of the shapelets that contributed to the classification, and the contribution rates in association with time-series data.

[0107] Fig. 9 is a diagram showing an example of a display screen of the second embodiment. Fig. 9 is an example of a display screen in which a display area 901 for displaying the contribution degree is added to the display screen of the first embodiment (Fig. 4). As shown in Fig. 9, by displaying the contribution degree, it becomes possible to grasp the distance component that contributes to classification for each shapelet. In the example of Fig. 9, it becomes possible to grasp that the difference due to the shape of the waveform is more effective for classification than the difference in the mean and variance.

[0108] In this way, the information processing device of the second embodiment can identify and visualize the contribution of each distance component, such as the deviation in shape, the deviation in the height direction (average) of the sensor value, and the deviation in the magnitude of the amplitude (variance). This allows the user to use the contribution as a criterion for determining the basis for classification.

[0109] As described above, according to the first and second embodiments, it is possible to improve the accuracy of estimation using an estimation model for time-series data.

[0110] Next, a hardware configuration of the information processing device according to the first or second embodiment will be described with reference to Fig. 10. Fig. 10 is an explanatory diagram showing an example of the hardware configuration of the information processing device according to the first or second embodiment.

[0111] The information processing device of the first or second embodiment includes a control device such as a CPU 51, a storage device such as a ROM (Read Only Memory) 52 or a RAM 53, a communication I / F 54 that connects to a network and communicates, and a bus 61 that connects each part.

[0112] The programs executed by the information processing device of the first or second embodiment are provided in a state that they are pre-installed in the ROM 52 or the like.

[0113] The program executed by the information processing device of the first or second embodiment may be configured to be provided as a computer program product by being recorded in an installable or executable format on a computer-readable recording medium such as a CD-ROM (Compact Disk Read Only Memory), a flexible disk (FD), a CD-R (Compact Disk Recordable), or a DVD (Digital Versatile Disk).

[0114] Furthermore, the program executed by the information processing device of the first or second embodiment may be configured to be stored on a computer connected to a network such as the Internet and provided by being downloaded via the network. Also, the program executed by the information processing device of the first or second embodiment may be configured to be provided or distributed via a network such as the Internet.

[0115] The program executed by the information processing device of the first or second embodiment can cause the computer to function as each part of the information processing device described above. In this computer, the CPU 51 can read the program from a computer-readable storage medium onto the main storage device and execute the program.

[0116] A configuration example of the embodiment will be described below. (Configuration example 1) adjusting a first distance between the input first time series data and a plurality of discrimination patterns each corresponding to any one of a plurality of classes using an existence degree that indicates a degree to which a plurality of time points in the first time series data are times at which any one of the plurality of discrimination patterns exists; calculating a first feature amount representing a feature of the first time series data by using the adjusted first distance; a model that inputs a second feature amount representing a feature of second time series data for learning and estimates the class corresponding to a discrimination pattern similar to the second time series data among a plurality of discrimination patterns, the discrimination pattern similar to the first time series data, the class, and the degree of presence being a time when a similar discrimination pattern exists, so as to reduce an estimation error, by inputting the first feature amount into an estimation model in which parameters are learned together with the similar discrimination pattern and the degree of presence of the time when a similar discrimination pattern exists. Processing section, An information processing device comprising: (Configuration example 2) The processing unit includes: executing a learning process for updating parameters of the estimation model, the plurality of discrimination patterns, and the presence degree, by using the second time-series data, so as to reduce an error between the class estimated by the estimation model and a correct class; The information processing device according to configuration example 1. (Configuration example 3) The processing unit includes: updating the degree of presence so that the values ​​of the degree of presence at adjacent times become closer to each other; The information processing device according to configuration example 2. (Configuration Example 4) The processing unit includes: adjusting the first distance by multiplying the abundance; The information processing device according to any one of configuration examples 1 to 3. (Configuration Example 5) the first distance is a sum of distances between the first time series data and one or more of the plurality of identification patterns; The information processing device according to any one of configuration examples 1 to 4. (Configuration Example 6) The processing unit includes: displaying, on a display device, the class estimated for the first feature amount, the discrimination pattern corresponding to the estimated class, and the degree of presence. 6. The information processing device according to any one of configuration examples 1 to 5. (Configuration Example 7) the first distance is calculated based on a plurality of types of distances between the first time-series data and the plurality of identification patterns and a contribution degree of each of the plurality of types of distances; The estimation model learns parameters including the similar discrimination pattern, the presence degree of the time when the similar discrimination pattern exists, and the contribution degree so as to reduce an estimation error. The information processing device according to any one of configuration examples 1 to 6. (Configuration Example 8) The plurality of types of distances are a difference between a statistic of partial time series data corresponding to the length of the identification pattern included in the first time series data and a statistic of the identification pattern; the distance between the shape of the partial time series data and the shape of the identification pattern; The distance between the shape of the standardized partial time series data and the shape of the identification pattern, A distance between a frequency component of the partial time series data and a frequency component of the identification pattern, A distance between a p-th order (p is an integer equal to or greater than 1) differential of the partial time series data and a p-th order differential of the identification pattern; Two or more of the following are included: The information processing device according to configuration example 7. (Configuration Example 9) The processing unit includes: displaying, on a display device, the class estimated for the first feature amount, the discrimination pattern corresponding to the estimated class, the degree of presence, and the degree of contribution. The information processing device according to configuration example 7. (Configuration Example 10) The processing unit includes: a distance calculation unit that adjusts the first distance using the degree of presence; A feature calculation unit that calculates the first feature amount; an estimation unit that obtains the classification pattern, the class, and the degree of presence similar to the first time-series data by inputting the first feature amount into the estimation model; Equipped with The information processing device according to any one of configuration examples 1 to 9. (Configuration Example 11) An information processing method executed by an information processing device, adjusting a first distance between input first time series data and a plurality of discrimination patterns each corresponding to any of a plurality of classes, using an existence degree that indicates a degree to which a plurality of time points in the first time series data are times at which any of the plurality of discrimination patterns exists; calculating a first feature amount representing a feature of the first time series data by using the adjusted first distance; a step of inputting a second feature amount representing a feature of second time series data for learning, and estimating the class corresponding to a discrimination pattern similar to the second time series data among a plurality of discrimination patterns, the discrimination pattern being similar to the second time series data, by inputting the first feature amount into an estimation model in which parameters are learned together with the similar discrimination pattern and the presence degree of a time at which a similar discrimination pattern exists so as to reduce an estimation error, thereby obtaining the discrimination pattern similar to the first time series data, the class, and the presence degree; An information processing method comprising: (Configuration Example 12) On the computer, adjusting a first distance between input first time series data and a plurality of discrimination patterns each corresponding to any of a plurality of classes, using an existence degree that indicates a degree to which a plurality of time points in the first time series data are times at which any of the plurality of discrimination patterns exists; calculating a first feature amount representing a feature of the first time series data by using the adjusted first distance; a step of inputting a second feature amount representing a feature of second time series data for learning, and estimating the class corresponding to a discrimination pattern similar to the second time series data among a plurality of discrimination patterns, the discrimination pattern being similar to the second time series data, by inputting the first feature amount into an estimation model in which parameters are learned together with the similar discrimination pattern and the presence degree of a time at which a similar discrimination pattern exists so as to reduce an estimation error, thereby obtaining the discrimination pattern similar to the first time series data, the class, and the presence degree; A program for executing the above.

[0117] Although some embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their modifications are included in the scope and spirit of the invention, and are included in the scope of the invention and its equivalents described in the claims. [Explanation of symbols]

[0118] 100, 100-2 Information processing device 101 Reception 102, 102-2 Output control section 111, 111-2 Distance calculation unit 112, 112-2 Feature calculation unit 113, 113-2 Learning Department 114 Estimation Department 131 Storage section 132 Display section

Claims

1. adjusting a first distance between the input first time series data and a plurality of discrimination patterns each corresponding to any of a plurality of classes using a degree of presence that indicates a degree to which a plurality of time points in the first time series data are times at which any of the plurality of discrimination patterns exists; calculating a first feature amount representing a feature of the first time series data by using the adjusted first distance; a model that inputs a second feature amount representing a feature of second time series data for learning and estimates the class corresponding to a discrimination pattern similar to the second time series data among a plurality of discrimination patterns, the discrimination pattern similar to the first time series data, the class, and the presence degree are obtained by inputting the first feature amount into an estimation model in which parameters are learned together with the similar discrimination pattern and the presence degree of a time at which a similar discrimination pattern exists so as to reduce an estimation error; Processing section, An information processing device comprising:

2. The processing unit includes: executing a learning process for updating parameters of the estimation model, the plurality of discrimination patterns, and the presence degree, by using the second time-series data, so as to reduce an error between the class estimated by the estimation model and a correct class; The information processing device according to claim 1 .

3. The processing unit includes: updating the degree of presence so that the values ​​of the degree of presence at adjacent times become closer to each other; The information processing device according to claim 2 .

4. The processing unit includes: adjusting the first distance by multiplying the degree of presence; The information processing device according to claim 1 .

5. the first distance is a sum of distances between the first time series data and one or more of the plurality of identification patterns; The information processing device according to claim 1 .

6. The processing unit includes: displaying, on a display device, the class estimated for the first feature amount, the discrimination pattern corresponding to the estimated class, and the degree of presence. The information processing device according to claim 1 .

7. the first distance is calculated based on a plurality of types of distances between the first time-series data and the plurality of identification patterns and a contribution degree of each of the plurality of types of distances; The estimation model learns parameters including the similar discrimination pattern, the presence degree of the time when the similar discrimination pattern exists, and the contribution degree so as to reduce an estimation error. The information processing device according to claim 1 .

8. The plurality of types of distances are a difference between a statistic of partial time series data corresponding to the length of the identification pattern included in the first time series data and a statistic of the identification pattern; the distance between the shape of the partial time series data and the shape of the identification pattern; The distance between the shape of the standardized partial time series data and the shape of the identification pattern, A distance between a frequency component of the partial time series data and a frequency component of the identification pattern, A distance between a p-th order (p is an integer equal to or greater than 1) differential of the partial time series data and a p-th order differential of the identification pattern, Two or more of the following are included: The information processing device according to claim 7.

9. The processing unit includes: displaying, on a display device, the class estimated for the first feature amount, the discrimination pattern corresponding to the estimated class, the degree of presence, and the degree of contribution. The information processing device according to claim 7.

10. The processing unit includes: a distance calculation unit that adjusts the first distance using the degree of presence; A feature calculation unit that calculates the first feature amount; an estimation unit that obtains the classification pattern, the class, and the degree of presence similar to the first time-series data by inputting the first feature amount into the estimation model; Equipped with The information processing device according to claim 1 .

11. An information processing method executed by an information processing device, adjusting a first distance between input first time series data and a plurality of discrimination patterns each corresponding to any of a plurality of classes, using a degree of presence that indicates a degree to which a plurality of time points in the first time series data are times at which any of the plurality of discrimination patterns exists; calculating a first feature amount representing a feature of the first time series data by using the adjusted first distance; a step of inputting a second feature amount representing a feature of second time series data for learning, and estimating the class corresponding to a discrimination pattern similar to the second time series data among a plurality of discrimination patterns, the discrimination pattern being similar to the second time series data, the first feature amount being inputted into an estimation model in which parameters are learned together with the similar discrimination pattern and the presence degree of a time at which a similar discrimination pattern exists, so as to reduce an estimation error, thereby obtaining the discrimination pattern similar to the first time series data, the class, and the presence degree; An information processing method comprising:

12. On the computer, adjusting a first distance between input first time series data and a plurality of discrimination patterns each corresponding to any of a plurality of classes, using a degree of presence that indicates a degree to which a plurality of time points in the first time series data are times at which any of the plurality of discrimination patterns exists; calculating a first feature amount representing a feature of the first time series data by using the adjusted first distance; a step of inputting a second feature amount representing a feature of second time series data for learning, and estimating the class corresponding to a discrimination pattern similar to the second time series data among a plurality of discrimination patterns, the discrimination pattern being similar to the second time series data, the first feature amount being inputted into an estimation model in which parameters are learned together with the similar discrimination pattern and the presence degree of a time at which a similar discrimination pattern exists, so as to reduce an estimation error, thereby obtaining the discrimination pattern similar to the first time series data, the class, and the presence degree; A program for executing the above.

Images