System, method, and program for predicting information
A system includes a learning object storing section that stores objects to be learned, a learning result storing section that stores learning results, and a control section connected to an input section. The control section computes a principal component coefficient vector of a first feature vector of an object to be processed that is designated by the input section, computes a principal component coefficient vector of a second feature vector using a principal component basis vector stored in the learning result storing section, and computes the second feature vector of the object to be processed using the principal component coefficient vector of the second feature vector.
1. A system configured to consistently and accurately predict information, comprising:
a learning object storage device that stores objects to be learned;
a learning result storage device that stores learning results; and
a control section including a processor and connected to an input section, configured to perform:
learning processing in which the control section:
computes a first feature vector and a second feature vector for each object to be learned stored in the learning object storage device,
generates a first normalized row vector for each object to be learned by subtracting an average vector of the first feature vectors from the first feature vectors to produce a set of first values and dividing each first value by an index that represents variation of the first feature vectors,
generates a second normalized row vector to be coupled with the first normalized row vector for each object to be learned by subtracting an average vector of the second feature vectors from the second feature vectors to produce a set of second values and dividing each second value by an index that represents variation of the second feature vectors, thereby equalizing scales of the variations of the principal component coefficients of the first and second feature vectors prior to principal component analysis,
uses coupling learning to generate a coupled principal component basis vector after generating the first and second normalized row vectors by:
generating a coupled row vector for each object to be learned by coupling the first normalized row vector and the second normalized row vector, and then
performing principal component analysis using the coupled row vectors of all of the objects to be learned, so as to generate a coupled principal component basis vector, and
stores the coupled principal component basis vector in the learning result storage device, together with an average vector of the coupled row vectors, and
prediction processing, in which the control section
computes a first feature vector of an object to be processed that is designated by the input section and computes a principal component coefficient of the first feature vector of the object to be processed, and
using the principal component coefficient of the first feature vector, the average vector of the coupled row vectors that is stored in the learning result storing device, and the coupled principal component basis vector that is stored in the learning result storing device, and computes a second feature vector of the object to be processed.
2. The system according to claim 1 , wherein the control section is configured to orthogonalize a set of components of the first feature vector in the coupled principal component basis vector and to store the orthogonalized vector in the learning result storage device.
3. The system according to claim 1 , wherein the control section is configured to perform post-processing, in which the control section
converts the computed second feature vector,
computes a difference between the converted second feature vector and the first feature vector of the object to be processed,
compares the difference with a reference value, and
when the difference is greater than the reference value, converts the difference and subtracts the converted difference from the second feature vector.
4. The system according to claim 3 , wherein the control section is configured to perform post-processing, in which the control section
compresses the computed second feature vector,
computes a difference between the compressed second feature vector and the first feature vector of the object to be processed,
compares the difference with a reference value, and
when the difference is greater than the reference value, increases the difference and subtracts the increased difference from the second feature vector.
5. The system according to claim 1 , wherein the index that represents the variation is a standard deviation of the principal component coefficient vectors.
6. A method for consistently and accurately predicting information performed by an information predicting computer system that includes
a learning object storage device that stores objects to be learned,
a learning result storage device that stores learning results, and
a control section including a processor and connected to an input section,
the method causes the control section to perform:
learning processing in which the control section
computes a first feature vector and a second feature vector for each object to be learned stored in the learning object storage device,
generates a first normalized row vector for each object to be learned by subtracting an average vector of the first feature vectors from the first feature vectors to produce a set of first values and dividing each first value by an index that represents variation of the first feature vectors,
generates a second normalized row vector to be coupled with the first normalized row vector for each object to be learned by subtracting an average vector of the second feature vectors from the second feature vectors to produce a set of second values and dividing each second value by an index that represents variation of the second feature vectors, thereby equalizing scales of the variations of the principal component coefficients of the first and second feature vectors prior to principal component analysis,
uses coupling learning to generate a coupled principal component basis vector after generating the first and second normalized row vectors by:
generating a coupled row vector for each object to be learned by coupling the first row vector and the second row vector, and then
performing principal component analysis using the coupled row vectors of all of the objects to be learned, so as to generate a coupled principal component basis vector, and
stores the coupled principal component basis vector in the learning result storage device, together with an average vector of the coupled row vectors, and
prediction processing, in which the control section
computes a first feature vector of an object to be processed that is designated by the input section, and computes a principal component coefficient of the first feature of the object to be processed, and
using the principal component coefficient of the first feature vector, the average vector of the coupled row vectors that is stored in the learning result storage device, and the coupled principal component basis vector that is stored in the learning result storage device, and computes a second feature vector of the object to be processed.
7. A non-transitory computer readable storage medium storing a program for consistently and accurately predicting information executed by an information predicting computer system that includes
a learning object storage device that stores objects to be learned,
a learning result storage device that stores learning results, and
a control section including a processor connected to an input section,
when the program is executed, the control section performs
learning processing in which the control section
computes a first feature vector and a second feature vector for each object to be learned stored in the learning object storage device,
generates a first normalized row vector for each object to be learned by subtracting an average vector of the first feature vectors from the first feature vectors to produce a set of first values and dividing each first value by an index that represents variation of the first feature vectors,
generates a second normalized row vector to be coupled with the first normalized row vector for each object to be learned by subtracting an average vector of the second feature vectors from the second feature vectors to produce a set of second values and dividing each second value by an index that represents variation of the second feature vectors, thereby equalizing scales of the variations of the principal component coefficients of the first and second feature vectors prior to principal component analysis,
uses coupling learning to generate a coupled principal component basis vector after generating the first and second normalized row vectors by:
generating a coupled row vector for each object to be learned by coupling the first normalized row vector and the second normalized row vector, and
performing principal component analysis using the coupled row vectors of all of the objects to be learned, so as to generate a coupled principal component basis vector, and
stores the coupled principal component basis vector in the learning result storage device, together with an average vector of the coupled row vectors, and
prediction processing, in which the control section
computes a first feature vector of an object to be processed that is designated by the input section, and computes a principal component coefficient of the first feature vector of the object to be processed, and
using the principal component coefficient of the first feature vector, the average vector of the coupled row vectors that is stored in the learning result storage device, and the coupled principal component basis vector that is stored in the learning result storage device, and computes a second feature vector of the object to be processed.