Change-point detection apparatus, method and program therefor
View Patent ↗A time-series model learning unit reads in time-series data sequentially, learns the parameters of a time-series model and stores the parameters in a storage device. A loss function calculating unit reads in sequentially from storage each item of the input time-series data one at a time and calculates values of a loss function. A complexity calculating unit sums the loss values to sequentially calculate complexity as fitting error resulting when a time-series model is fit to the input data. Complexity is stored. A change-point searching unit reads in complexity regarding time-series data before and after change-point candidates from storage with respect to all change-point candidates, compares the sum of the complexities with complexity regarding all time-series data in a case where a change point is assumed to be absent, calculates the change-point score based upon the difference between the compared values, and detects and outputs the change point.
1. A change-point detection apparatus for reading in time-series data sequentially as an input and detecting a change point which is a time point where the input time-series data changes suddenly, comprising:
a time-series model learning unit for learning parameters of a time-series model sequentially while the time-series data is being read in;
a storage device for storing the parameters of the time-series model, which have been learned by the time-series model learning unit, and complexity as fitting error that results when the time-senes model is fit to the input time-series data;
a loss-function calculating unit for reading the pararneters from said storage device and calculating loss of the time-series model with regard to each item of the input time-series data one at a time;
a complexity calculating unit for reading parameters and complexity from said storage device, calculating complexity of a partial series of the input time-series data sequentially while updating the same, and updating content stored in the storage device; and
a change-point searching unit for reading complexity with regard to time-series data before and after change-point candidates from the storage device with respect to all change-point candidates of the input time-series data and comparing the sum of the complexities before and after the change-time candidates with complexity when a change point is assumed to be absent, thereby calculating change-point score and detecting a change point; wherein the change point detected by the change-point searching unit is output.
2. A change-point detection apparatus for reading in time-series data sequentially as an input and detecting a change point which is a time point where the input time-series data changes suddenly, comprising:
a time-series model learning unit for learning parameters of a time-series model sequentially while the time-series data is being read in;
a storage device for storing the parameters of the time-series model, which have been learned by said time-series model learning unit, and complexity as fitting error that results when the time-senes model is fit to the input time-series data;
a loss-function calculating unit for reading parameters from said storage device and calculating loss of the time-series model with regard to each item of the input time-series data one at a time;
a complexity calculating unit for sequentially adding a loss value calculated by said loss-function calculating unit to complexity read in from said storage device, thereby calculating complexity as fitting error that results when the time-series model is fit to a partial series of the input time-series data, and updating content stored in said storage device; and
a change-point determination unit for reading complexity with regard to time-series data before and after change-point candidates from the storage device with respect to all change-point candidates of the input time-series data and comparing the sum of the complexities before and after the change-time candidates with complexity when a change point is assumed to be absent, thereby calculating change-point score and detecting a change point; wherein the change point detected by the change-point determination unit is output.
3. A change-point detection method for reading in time-series data sequentially as an input and detecting a change point which is a time point where the input time-series data changes suddenly, comprising:
a step of reading in time-series data sequentially;
a step of learning parameters of a time-series model sequentially from the input time-series data;
a loss-function calculating step of calculating loss of the time-series model with regard to each item of the input time-series data one at a time;
a complexity calculating step of sequentially calculating complexity as fitting error that results when the time-series model is fit to the input time-series data; and
a change-point searching step of calculating complexity with regard to time-series data before and after change-point candidates with respect to all change-point candidates of the input time-series data and comparing the sum of the complexities before and after the change-time candidates with complexity when a change point is assumed to be absent, thereby calculating change-point score and detecting a change point; wherein the change point detected at said change-point search step is output.
4. A change-point detection method for reading in time-series data sequentially as an input and detecting a change point which is a time point where the input time-series data changes suddenly, comprising:
a step of reading in time-series data sequentially;
a step of learning parameters of a time-series model sequentially from the input time-series data;
a loss-function calculating step of calculating loss of the time-series model with regard to each item of the input time-series data one at a time;
a complexity calculating step of sequentially adding a loss value calculated at said loss-function calculating step, thereby calculating complexity as a fitting error that results when the time-series model is fit to a partial series of the input time-series data; and
a change-point determination step of reading, from a storage device, complexity with regard to time-series data before and after change-point candidates with respect to all change-point candidates of the input time-series data and comparing the sum of the complexities before and after the change-time candidates with complexity when a change point is assumed to be absent, thereby calculating change-point score and detecting a change point; wherein the change point detected at said change-point determination step is output.
5. A computer program product for enabling a computer system to carry out a computer-assisted method for reading in time-series data sequentially as an input and detecting a change point, comprising:
software instructions for the computer system, and
a computer readable medium including the software instructions;
wherein the software instructions define operations, comprising:
reading time-senes data seciuentially,
learning parameters of a time-series model sequentially while the time-series data is being read in;
storing the parameters of the time-series model, which have been learned by said time-series model learning means, and complexity as fitting error that results when the time-series model is fit to the input time-series data;
reading parameters from said storage device and calculating loss of the time-series model with regard to each item of the input time-series data one at a time;
reading parameters and complexity from said storage device, calculating complexity of a partial series of the input time-series data sequentially while updating the same, and updating content stored in said storage device; and
reading complexity with regard to time-series data before and after change-point candidates from the storage device with respect to all change-point candidates of the input time-series data and comparing the sum of the complexities before and after the change-time candidates with complexity when a change point is assumed to be absent, thereby calculating change-point score and detecting a change point.
6. A computer program product for enabling a computer system to carry out a computer-assisted method for reading in time-series data sequentially as an input and detecting a change point, comprising:
software instructions for the computer system, and
a computer readable medium including the software instructions;
wherein the software instructions define operations, comprising:
reading time-series data sequentially,
learning parameters of a time-series model sequentially while the time-series data is being read in;
storing the parameters of the time-series model, which have been leamed by said time-series model leaming means, and complexity as fitting error that results when the time-series model is fit to the input time-series data;
reading parameters from said storage device and calculating loss of the time-series model with regard to each item of the input time-series data one at a time;
sequentially adding a loss value calculated by said loss-function calculating means to complexity read in from said storage device, thereby calculating complexity as fitting error that results when the time-series model is fit to a partial series of the input time-series data, and updating content stored in said storage device; and
reading complexity with regard to time-series data before and after change-point candidates from said storage device with respect to all change-point candidates of the input time-series data and comparing the sum of the complexities before and after the change-time candidates with complexity when a change point is assumed to be absent, thereby calculating change-point score and detecting a change point.