JP7708031B2 - Information processing device, information processing method, and program - Google Patents

Description

The present disclosure relates to an information processing device, an information processing method, and a program, and in particular to an information processing device, an information processing method, and a program that calculates a predicted value.

In technical fields such as causal analysis, which analyzes causal relationships between data, a process of calculating a predicted value (estimated value) is sometimes performed based on the relative distance between data. In relation to this, the causal relationship learning device disclosed in Patent Document 1 includes a correlation determination unit that determines the correlation between measurements taken by two sensors, and an estimation unit that determines the causal relationship between the two sensors by estimating the cause measurement value from the result measurement value when the correlation is lower than a predetermined standard. Furthermore, as a related technique, Non-Patent Document 1 discloses causal analysis using the NPMR (Non-Parametric Multiplicative Regression) model.

International Publication No. 2018/198267

Nicolaou N et al., “A Nonlinear Causality Estimator Based on Non-Parametric Multiplicative Regression”, Frontiers in Neuroinformatics, Volume 10, June 2016.

The prediction formula used in Patent Document 1 has a relatively simple formula for calculating weights based on the relative distance between data, and therefore has poor expressive power as a nonlinear model. To solve this problem, for example, a prediction formula with an increased number of weights could be used, or a multiplication operation could be used to calculate the weights in the prediction formula. However, when such a prediction formula is used, the amount of calculation related to the weights increases, and a long calculation time is required.

The present disclosure has been made against the background of the above circumstances, and aims to provide an information processing device, an information processing method, and a program that can achieve high-speed processing in prediction processing using a prediction formula that includes calculation of the relative distance between data.

One aspect of the present disclosure for achieving the above object is an information processing device that includes a data acquisition unit that acquires explanatory variable data expressed as a matrix of T (where T is an integer of 2 or more) rows and m (where m is an integer of 1 or more) columns, and objective variable data expressed as a vector having T components, and a prediction processing unit that calculates a predicted value for the objective variable using a predetermined prediction formula including a calculation of a relative distance between data of a row of interest and data of each of the other rows of the explanatory variable data, and the data acquired by the data acquisition unit, and the prediction processing unit calculates the predicted value using the prediction formula in which the relative distance with each of the other rows is expressed by a vector, or by thinning out data of rows of the other rows whose relative distance is equal to or greater than a predetermined threshold.

In addition, another aspect of the present disclosure for achieving the above object is an information processing method in which an information processing device acquires explanatory variable data expressed as a matrix with T (where T is an integer of 2 or more) rows and m (where m is an integer of 1 or more) columns, and objective variable data expressed as a vector having T components, and calculates a predicted value for the objective variable using a predetermined prediction formula including a calculation of a relative distance between data of a row of interest and data of each of the other rows of the explanatory variable data, and the acquired data, and in the step of calculating the predicted value, the predicted value is calculated using the prediction formula in which the relative distance with each of the other rows is expressed by a vector, or by thinning out data of rows among the other rows whose relative distance is equal to or greater than a predetermined threshold.

In addition, another aspect of the present disclosure for achieving the above object is a program that causes a computer to execute a data acquisition step of acquiring explanatory variable data expressed as a matrix of T (where T is an integer of 2 or more) rows and m (where m is an integer of 1 or more) columns and objective variable data expressed as a vector having T components, and a prediction processing step of calculating a predicted value for the objective variable using a predetermined prediction formula including a calculation of a relative distance between data of a row of interest and data of each of the other rows among the explanatory variable data, and the data acquired in the data acquisition step, and in which the prediction processing step calculates the predicted value using the prediction formula in which the relative distance with each of the other rows is expressed by a vector, or by thinning out data of each of the other rows in which the relative distance is equal to or greater than a predetermined threshold.

According to the present disclosure, it is possible to provide an information processing device, an information processing method, and a program that can achieve high-speed processing in prediction processing using a prediction formula that includes calculation of the relative distance between data.

1 is a block diagram showing an example of a configuration of an information processing device according to an embodiment; FIG. 1 is a diagram showing an example of a source code of a program when the calculation represented by the original NPMR prediction formula is implemented by a computer program. FIG. 13 is a diagram showing an example of source code of a computer program that implements the calculations represented by the vector prediction formula. FIG. 13 is a diagram showing an example of a recurrence plot. 1 is a block diagram illustrating an example of a hardware configuration of an information processing device according to an embodiment. 10 is a flowchart illustrating an example of an operation of the information processing device according to the embodiment. 11 is a graph showing experimental results in a first experiment. 13 is a graph showing experimental results in a second experiment. 13 is a graph showing experimental results in a third experiment.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
1 is a block diagram showing an example of a configuration of an information processing device 10 according to an embodiment. The information processing device 10 performs a process of predicting data of a response variable using data of explanatory variables and data of a response variable in order to perform a causal analysis for analyzing a causal relationship between data of explanatory variables and data of a response variable, for example, but may perform a prediction process described later for any other purpose. As an example, the information processing device 10 has a data acquisition unit 101 and a prediction processing unit 102 as shown in FIG. 1.

The data acquisition unit 101 accepts input of data to be used in the prediction process. The data acquisition unit 101 accepts input of a data series of explanatory variables and a data series of objective variables. In this embodiment, both the explanatory variable data and the objective variable data are time-series data, but they do not necessarily have to be time-series data. In particular, the data acquisition unit 101 acquires explanatory variable data expressed as a matrix with T rows and m columns, and objective variable data expressed as a vector having T components.

Specifically, the data acquisition unit 101 acquires data on explanatory variables shown in the following formula (1).

<Formula (1)>

The data acquisition unit 101 also acquires data on the objective variable shown in the following equation (2).

<Formula (2)>

Here, T is an integer of 2 or more, and m is an integer of 1 or more. Note that m represents the number of types of data used as explanatory variables. In other words, m predictors are used for the prediction process. Also, T is the number of data in each data series (each time series data). To give a specific example, for example, the data series in the first column of X is time series data of temperature, the data series in the second column is time series data of precipitation, and the data series indicated by Y is time series data of electricity usage. In this case, the prediction processing unit 102 described later calculates a predicted value of electricity usage based on the time series data of temperature, the time series data of precipitation, and the time series data of electricity usage. Note that these are merely examples, and the specific data of each data series is not limited to the above-mentioned examples.

The data acquisition unit 101 may set explanatory variable data represented as a matrix of T rows and m columns and target variable data represented as a vector having T components from the input data series. That is, the data acquisition unit 101 may accept inputs specifying the values of x _i,j (where i is an integer between 1 and T, and j is an integer between 1 and m) which are components of the matrix X, and y _t (where t is an integer between 1 and T) which are components of the vector Y, or may set these values from the input data according to user instructions. The data acquisition unit 101 may also normalize the data in advance so that the standard deviation σ _j described later is a predetermined value (for example, 1).

The prediction processing unit 102 calculates a predicted value for the objective variable using a predetermined prediction formula including a calculation of the relative distance (difference) between the data of a row of interest (data of row t) and each of the other rows of data (data of each row other than row t) among the explanatory variable data, and the data (X and Y) acquired by the data acquisition unit 101. More specifically, the predetermined prediction formula is a prediction formula including a weight calculation based on the above-mentioned relative distance. Note that in this predetermined prediction formula, the weight is set so that the greater the relative distance, the smaller the weight for the objective variable value.

In this embodiment, the prediction processing unit 102 calculates a predicted value using, as the above-mentioned predetermined prediction formula, specifically, for example, a prediction formula based on the prediction formula of the NPMR model disclosed in Non-Patent Document 1 (a prediction formula obtained by modifying the prediction formula of the NPMR model). Here, the prediction formula of the NPMR model disclosed in Non-Patent Document 1 is expressed as the following formula (3). Here, in the prediction formula of formula (3), the weight w _i,j applied to the value of the objective variable is expressed by formula (4). Note that σ _j is the standard deviation of the data in the jth column of the explanatory variable X. In addition, in the present disclosure, a variable representing a predicted value appearing on the left side of formula (3) may be expressed as y _t ^.

<Formula (3)>

<Formula (4)>

In this way, in NPMR, the predicted value _yt ^ of _yt , which is the t-th component of the data series of the objective variable Y, is calculated by weighting each component other than the t-th component of the data series of the objective variable Y. As is clear from the above prediction formula, in the NPMR model, when all components of the objective variable Y, i.e., _y1 to _yT , are real data (i.e., when none of them are null values), all of _y1 ^ to _yT ^ can be calculated by setting 1 to T as the value of t. Therefore, the accuracy of the prediction using the explanatory variable X can be known by comparing each of the real data _y1 to _yT with each of the predicted values _y1 ^ to _yT ^. On the other hand, yt _, which is the t-th component of the data series of the objective variable Y, is not used to calculate the predicted value yt _^ . If only one value (for example, unknown _yt ) of _y1 ^ to _yT ^ is to be calculated, all of the T components of the objective variable Y do not necessarily have to be real data. That is, the data acquisition unit 101 may acquire a vector Y including one component that is a null value (missing value) as a data string of the objective variable Y. In other words, a predetermined t-th component of a vector having T components acquired by the data acquisition unit 101 may be a null value.

As can be seen from formula (4), the weight is calculated by calculating the relative distance (difference) between the data of the row of interest (data of row t) and each of the other rows (data of each row other than row t) among the data of the explanatory variables. Although the NPMR model can adequately express nonlinearity compared to the model shown in Patent Document 1 and the like, the amount of calculation related to the weight is large, and the processing time required for calculating the predicted value is long. Therefore, in this embodiment, in order to realize high-speed processing, the prediction processing unit 102 uses a prediction formula in which the relative distance is expressed by a vector, and calculates the predicted value by thinning out the data of the row whose relative distance is equal to or greater than a predetermined threshold. In other words, the calculation of the predicted value by the prediction processing unit 102 has the characteristics of vectorization of the relative distance and thinning out the data. Note that this embodiment has both of these characteristics, but the prediction processing unit 102 may calculate the predicted value by adopting only one of the characteristics.

First, the vectorization of relative distance will be described. The inventors have found that the prediction formula of the NPMR model can be transformed into the following prediction formula (formula (5)) in which the relative distance is represented by a vector. In this disclosure, such a transformed prediction formula will be referred to as a vectorized prediction formula.

<Formula (5)>

Here, the subscript -t means that the t-th (t-th row) component is removed. For example, for a vector y shown in the following equation (6), y _-t is defined as shown in the following equation (6).

<Formula (6)>

Therefore, in equation (5), vector y _−t is a vector having T−1 components obtained by excluding the t-th component from the data string of the response variable expressed as a vector having T components.

In addition, in formula (5), the vector Δx _-t is a vector having T-1 components whose components are the relative distance between the dataset of the row other than the t row in the explanatory variable data and the dataset of the t row. In other words, the vector Δx _-t is a vector having T-1 components obtained by removing the t-th component from T vectors whose i-th component is the relative distance between the dataset of the i row and the dataset of the t row in the explanatory variable data. The relative distance between the dataset of the row other than the t row in the explanatory variable data and the dataset of the t row can also be said to be the relative distance between the vector represented by the m components (x _i,1 , x _i,2 , ..., x _i,m ) of the i row other than the t row in the explanatory variable data and the vector represented by the m components (x _t,1 , x _t,2 , ..., x _t,m ) of the t row in the explanatory variable data.

As a simple example, when m=1, the vector Δx _−t can be expressed as follows:

<Formula (7)>

In addition, in the formulas shown in this disclosure, such as formula (5), the operator "." is an operator indicating an inner product (inner product operator), and the operator "o" is an operator indicating a Hadamard product (Hadamard product operator). In addition, in the formulas shown in this disclosure, such as formula (5), the function "exp" is an exponential function that acts on each component of a vector. In other words, the function "Exp" means that the function "exp" is applied to each component.

2 is a diagram showing an example of a source code of a program when the calculation shown in the formula (3), which is the original NPMR prediction formula, is implemented by a computer program. FIG. 3 is a diagram showing an example of a source code of a program when the calculation shown in the formula (5), which is the vectorized prediction formula, is implemented by a computer program. The source code shown in FIG. 2 and FIG. 3 shows an example when Julia is used as a programming language, but the calculation of the prediction formula may be implemented by other programming languages. As shown in FIG. 3, when the vectorized prediction formula is used, it is possible to implement it by a code that performs vector calculations, so that the predicted value can be calculated by a code that reduces the number of calculations by loop processing compared to the code that implements the calculation of the prediction formula before vectorization (the original NPMR prediction formula). Therefore, by using the vectorized prediction formula, high-speed processing can be realized in the prediction processing of the predicted value. The speed-up of processing when the vectorized prediction formula is used can be experimentally understood from the results shown in FIG. 7, which will be described later.

Next, data thinning will be described. As can be seen from formulas (3) and (4), in a prediction formula in which weights are calculated based on relative distances, such as the NPMR prediction formula, the component of the objective variable Y corresponding to a row having a large relative distance from the data of the t row of the explanatory variable X has a low degree of contribution in the calculation of the predicted value y _t ^. Therefore, in this embodiment, in order to further speed up the calculation process of the predicted value, the prediction processing unit 102 calculates the predicted value by thinning and calculating the data of the row whose relative distance is equal to or greater than a predetermined threshold. Specifically, the prediction processing unit 102 calculates the predicted value y _t ^ using the prediction formula shown in the following formula (8). Formula (8) is a prediction formula obtained by further modifying the vectorization prediction formula shown in formula (5). The predicted value y _t ^ calculated by formula (8) is an approximation of the predicted value calculated by the prediction formula shown in formula (3) or formula (5).

<Formula (8)>

In the above formula, Δx _−t ^rec is defined by the following formula (9), and y _−t ^rec is defined by the following formula (10).

<Formula (9)>

<Formula (10)>

Here, R(t) is a vector having T-1 components generated based on the relative distance between the data set of rows other than row t in the explanatory variable data and the data set of row t. R(t) is used as a filter for extracting components whose relative distance is less than a predetermined threshold, that is, components whose contribution in the calculation of the predicted value is equal to or greater than a predetermined standard. The prediction processing unit 102 generates R(t), which is a filter for extracting data of rows whose relative distance is less than a threshold, in other words, a filter for extracting components whose contribution in the calculation of the predicted value is equal to or greater than a predetermined standard, based on the relative distance, and thins out data of rows whose relative distance is equal to or greater than a threshold using the filter. R(t) corresponds to a point plotted in a recurrence plot (see FIG. 4) that represents the relative distance between the data of the explanatory variables, and is defined by the following formulas (11) and (12).

<Formula (11)>

<Formula (12)>

In addition, in formula (12), k = 1, 2, ..., t-1, t+1, ..., T. That is, k is an integer from 1 to T excluding t. In addition, in formula (12), d( _xk , _xt ) is the relative distance between the kth row of the data set and the tth row of the data set in the explanatory variable X. In addition, δ is the above-mentioned predetermined threshold value. As can be seen from the relationship between R(t) and the recurrence plot shown in FIG. 4, R(t) shown in formula (11) is a vector expressing the distribution of the recurrence plot at time t with 0 and 1.

As described above, the components of R(t) are 0 when the relative distance between the data set of the tth row in the data of the explanatory variable X is equal to or greater than the threshold δ, and are 1 when the relative distance is less than the threshold δ. Then, as shown in formula (9), Δx _-t ^rec is defined by the Hadamard product of R(t) having T-1 components that take the value of 0 or 1 and Δx _-t . That is, Δx ^rec _-t is a vector in which the values of the components of Δx _-t have been changed, and the values of each component corresponding to the row in the explanatory variable data whose relative distance between the data of the tth row and the data of the tth row has been changed to 0. In other words, Δx ^rec _-t is a vector in which the values of the components of Δx _-t have been changed, and it can be said that the values of the components of Δx _-t whose relative distance is equal to or greater than the threshold have been changed to 0. Similarly, as shown in formula (10), y _-t ^rec is defined by the Hadamard product of R(t) having T-1 components that take the value of 0 or 1 and y _-t . That is, y _-t ^rec is a vector in which the value of the component of y _-t is changed, and the value of each component corresponding to a row in which the relative distance from the data of the t row in the explanatory variable data is equal to or greater than a threshold is changed to 0. In other words, y _-t ^rec is a vector in which the value of the component of y _-t is changed to 0, and the value of the component of y _-t having the same index (element number) as the component whose value is changed to 0 in Δx ^rec _-t is changed to 0. Therefore, in contrast to the prediction formula shown in formula (5), in the prediction formula shown in formula (8), the prediction value is calculated using a vector in which the value of the component _whose relative distance is equal to or greater than a threshold, that is, the component that does not substantially contribute to the prediction value calculation, is changed to 0 among the components of Δx _-t , and a vector in which the value of the component that does not substantially contribute to the prediction value calculation is changed to 0 among the components of y -t. In this way, the prediction processing unit 102 performs calculations by thinning out data from rows whose relative distances are equal to or greater than a predetermined threshold, in other words, by excluding such data, it is possible to calculate a predicted value with a smaller amount of calculations, thereby realizing high-speed processing in the prediction process of the predicted value. In addition, memory consumption can be reduced accordingly.

Next, an example of the hardware configuration of the information processing device 10 will be described. FIG. 5 is a block diagram showing an example of the hardware configuration of the information processing device 10. As shown in FIG. 5, the information processing device 10 includes an input/output interface 151, a memory 152, and a processor 153.

The input/output interface 151 is an interface for communicatively connecting to other devices (e.g., an input device or an output device) as necessary. For example, the input/output interface 151 may be used by the data acquisition unit 101 to acquire data, or by the prediction processing unit 102 to output prediction results.

Memory 152 is composed of, for example, a combination of volatile memory and non-volatile memory. Memory 152 is used to store software (computer programs) including one or more instructions executed by processor 153, data used for various processes, etc.

1 by reading and executing software (computer programs) from the memory 152. The processor 153 may be, for example, a microprocessor, a microprocessor unit (MPU), or a central processing unit (CPU). The processor 153 may include multiple processors.
In this manner, the information processing device 10 has the functions of a computer.

The program includes instructions (or software code) that, when loaded into a computer, cause the computer to perform one or more functions described in the embodiments. The program may be stored on a non-transitory computer-readable medium or a tangible storage medium. By way of example and not limitation, computer-readable media or tangible storage media include random-access memory (RAM), read-only memory (ROM), flash memory, solid-state drive (SSD) or other memory technology, CD-ROM, digital versatile disc (DVD), Blu-ray (registered trademark) disk or other optical disk storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage device. The program may be transmitted on a transitory computer-readable medium or communication medium. By way of example and not limitation, a transitory computer-readable medium or communication medium includes electrical, optical, acoustic, or other forms of propagated signals.

Next, the processing of the information processing device 10 will be described with reference to a flowchart. FIG. 6 is a flowchart showing an example of the operation of the information processing device 10 according to this embodiment.

In step S100, the data acquisition unit 101 sets data on explanatory variables and target variables. In this embodiment, specifically, the data acquisition unit 101 sets data on explanatory variables expressed as a matrix with T rows and m columns, and data on target variables expressed as a vector having T components, to be used in calculating the predicted value. Next, in step S101, the prediction processing unit 102 calculates the predicted value using a prediction formula for speeding up. In this embodiment, the prediction processing unit 102 calculates the predicted value using a vectorized prediction formula that is modified so that calculation can be performed by thinning out data, such as the prediction formula shown in formula (8). However, the calculation of the predicted value by the prediction processing unit 102 may be performed by adopting only one of the features of vectorization of relative distance and the feature of thinning out data. Next, in step S102, the prediction processing unit 102 outputs the predicted value calculated in step S101, i.e., the prediction result. The prediction processing unit 102 may output to an output device such as a display, or may output to a storage device to store the prediction result.

The above describes the embodiment. According to the information processing device 10, the prediction processing unit 102 calculates a predicted value using a prediction formula in which the relative distance between the data of a row of interest and the data of each of the other rows of explanatory variable data is expressed by a vector, or by thinning out the data of the rows in which the relative distance is equal to or greater than a predetermined threshold value among the other rows. In the former case, the loop processing to be executed is reduced compared to the case where a prediction formula in which the relative distance is not expressed by a vector is calculated by a computer program, so that high-speed processing can be realized. In the latter case, data that does not substantially contribute to the calculation of the predicted value is thinned out and processing is performed, so that high-speed processing can be realized. In addition, memory consumption can also be reduced. In particular, the prediction processing unit 102 uses a prediction formula based on the prediction formula of the NPMR model as a prediction formula used to calculate the predicted value, so that prediction processing that sufficiently reflects nonlinearity can be realized at high speed.

Next, experimental results will be shown regarding high-speed processing by the prediction processing unit 102. The measurement environment used for this experiment is as follows.
Operating system: macOS Catalina
Processor: 2.8GHz, quad-core, Intel Core i7
Memory: 16GB, 2133Mhz, LPDDR3

In the first experiment, two time series data defined as follows were used as test data, and explanatory variable X and objective variable Y were set from these time series data. In addition, in formula (13), e ₁ (t) and e ₂ (t) are white noise. In the test data shown in formula (13), there is a nonlinear causality from x ₂ to x ₁ .

<Formula (13)>

In the second experiment, three time series data defined as follows were used as test data, and explanatory variable X and objective variable Y were set from these time series data. In addition, in formula (14), e ₁ (t), e ₂ (t), and e ₃ (t) are white noise. In the test data shown in formula (14), there is a nonlinear causal relationship from x ₁ to x ₂ , a nonlinear causal relationship from x ₁ to x ₃ , and a linear causal relationship from x ₂ to x ₃ .

<Formula (14)>

Figure 7 is a graph showing the results of the first experiment. Figure 8 is a graph showing the results of the second experiment. Figures 7 and 8 show a graph of the time required for analysis using the original NPMR (graph labeled "NPMR"), a graph of the time required for analysis using NPMR with relative distance vectorization (graph labeled "Fast NPMR-1"), and a graph of the time required for analysis using NPMR with a combination of relative distance vectorization and data thinning (graph labeled "Fast NPMR-2"). As shown in Figures 7 and 8, it can be seen that the technology shown in this embodiment is used to speed up processing. The speedup is particularly noticeable the more data there is and the more complex the analysis target is.

In addition, in the third experiment, a prediction formula other than the above-mentioned prediction formula was used to confirm the effect of thinning out the data. Here, a prediction formula using the kernel method was examined. Specifically, the prediction processing unit 102 performs processing using the prediction formula shown in the following formula (16) in order to quickly process the prediction formula shown in the following formula (15).

<Formula (15)>

<Formula (16)>

In formula (16), the data used in the calculation is thinned out using the above-mentioned filter, that is, a filter for extracting components whose relative distance is less than a predetermined threshold (components whose contribution in the calculation of the predicted value is equal to or greater than a predetermined standard). Here, R(t) in formula (16) is defined as follows, and represents a set of indexes (row numbers) indicating components whose relative distance is less than the threshold δ. In formula (17), D(t _i ) indicates the data of the i row of explanatory variable X set by the data acquisition unit 101, and D(t _i ) is calculated as the relative distance from the data D(t) of the t row of explanatory variable X, and is compared with the threshold.

<Formula (17)>

The prediction processing unit 102 generates R(t) and performs the calculation of the prediction formula shown in equation (16). Figure 9 is a graph showing the experimental results of the third experiment. Note that the test data used in the third experiment is the same as that in the first experiment. Figure 9 shows a graph of the time required for analysis without thinning out data based on relative distance (graph labeled "normal version") and a graph of the time required for analysis after thinning out data based on relative distance (graph labeled "fast version"). As shown in Figure 9, it can be seen that the processing speed is increased by thinning out data using a filter.

The present invention is not limited to the above-described embodiment, and can be modified as appropriate without departing from the spirit of the present invention. For example, in the above-described embodiment, the information processing device 10 has the functions of the data acquisition unit 101 and the prediction processing unit 102, but some or all of these functions may be implemented by another device (e.g., a server, etc.). In other words, the processing described in the above-described embodiment may be realized by a system composed of one or more devices.

10 Information processing device 101 Data acquisition unit 102 Prediction processing unit 151 Input/output interface 152 Memory 153 Processor

Claims (7)

a data acquisition unit that acquires data on explanatory variables represented as a matrix with T rows and m columns (where T is an integer of 2 or more) and data on objective variables represented as a vector having T components;
a prediction processing unit that calculates a predicted value for the objective variable using a predetermined prediction formula including a calculation of a relative distance between a row of data of interest and each of the other rows of data among the explanatory variable data, and the data acquired by the data acquisition unit;
The prediction processing unit calculates the predicted value using the prediction formula in which the relative distance to each of the other rows is expressed by a vector, or by thinning out data of rows among the other rows whose relative distance is equal to or greater than a predetermined threshold. The information processing device according to claim 1 , wherein the prediction processing unit generates a filter for extracting data of the row in which the relative distance is less than the threshold value based on the relative distance, and thins out data of the row in which the relative distance is equal to or greater than the threshold value using the filter. The information processing apparatus according to claim 1 , wherein the predetermined prediction formula is a prediction formula based on a prediction formula of a Non-Parametric Multiplicative Regression (NPMR) model. The information processing device according to claim 3 , wherein the prediction processing unit calculates the predicted value using the following formula as the prediction formula:
however,
In the following formula:
is the predicted value,
t corresponds to the row number of the row of interest,
Δx ^rec _-t is a vector obtained by changing the values of the components of a vector having T-1 components whose components are the relative distance between a dataset of rows other than row t in the explanatory variable data and the dataset of row t, and the values of each component corresponding to the row whose relative distance is equal to or greater than the threshold are changed to 0;
y ^rec _-t is a vector obtained by changing the values of the components of a vector having T-1 components excluding the t-th component from the objective variable represented as a vector having T components, in which the values of the components corresponding to the rows in which the relative distance is equal to or greater than the threshold value are changed to 0;
exp is the exponential function,
Exp is the exponential function acting on each component of the vector.
The information processing apparatus according to claim 1 , wherein the data of the explanatory variables and the data of the objective variables are time-series data. An information processing device,
Obtain data on explanatory variables represented as a matrix with T rows and m columns (where T is an integer of 2 or more) and data on objective variables represented as a vector having T components;
calculating a predicted value for the objective variable using a predetermined prediction formula including a calculation of a relative distance between the data of a row of interest and each of the other rows of data among the data of the explanatory variables, and the acquired data;
The information processing method, in which in the step of calculating the predicted value, the predicted value is calculated using the prediction formula in which the relative distance to each of the other rows is expressed by a vector, or by thinning out data of rows among the other rows whose relative distance is equal to or greater than a predetermined threshold. a data acquisition step of acquiring data on explanatory variables represented as a matrix with T rows and m columns (where T is an integer of 2 or more) and data on objective variables represented as a vector having T components;
a prediction processing step of calculating a predicted value for the objective variable using a predetermined prediction formula including a calculation of a relative distance between a row of data of interest and each of the other rows of data among the explanatory variable data, and the data acquired in the data acquisition step;
A program for calculating the predicted value in the prediction processing step by using the prediction formula in which the relative distance to each of the other rows is expressed by a vector, or by thinning out data of rows among the other rows whose relative distance is equal to or greater than a predetermined threshold.