Model training through target attribute exclusion

A classification method to be executed by a classification apparatus, the method comprising:

• receiving learning data for model training, wherein the learning data is associated with a plurality of attributes;
• selecting a target attribute which is an attribute of non-input target data that is to be excluded from model training;
• generating a plurality of new data combinations from data of the learning data with attributes other than the target attribute;
• generating a plurality of sets of initial values of model parameters by inputting random number seeds into a random function, wherein the random number seeds are selected such that the initial values have a uniform distribution within a predetermined range;
• dividing the learning data into a plurality of training data sets, wherein each training data set comprises fewer data records than a total number of data records in the learning data;
• generating a plurality of models, wherein each model of the plurality of models has a different set of initial values of model parameters;
• training each of the plurality of models using the plurality of training data sets;
• inputting each of the plurality of new data combinations into the plurality of models and evaluating a model accuracy for each of the plurality of models;
• selecting a model from the plurality of models based on the evaluated model accuracy of each of the plurality of models and performing further model training on the selected model, the further model training comprising:
• for each training epoch, generating a new random number seed to determine a learning order for the training data sets;
• and
• training the selected model using the training data sets in the determined learning order;
• and transmitting a control signal to assign execution of prediction processes using the selected model to a central processing unit responsive to the selected model performing two-class classification and to a graphics processing unit responsive to the selected model performing multi-class classification.

A non-transitory computer-readable storage medium having stored therein a classification program for causing a computer having a central processing unit and a graphics processing unit to execute:

• receiving learning data for model training, wherein the learning data is associated with a plurality of attributes;
• selecting a target attribute which is an attribute of non-input target data that is to be excluded from model training;
• generating a plurality of new data combinations from data of the learning data with attributes other than the target attribute;
• generating a plurality of sets of initial values of model parameters by inputting random number seeds into a random function, wherein the random number seeds are selected such that the initial values have a uniform distribution within a predetermined range;
• dividing the learning data into a plurality of training data sets, wherein each training data set comprises fewer data records than a total number of data records in the learning data;
• generating a plurality of models, wherein each model of the plurality of models has a different set of initial values of model parameters from the plurality of sets of initial values;
• training each of the plurality of models using the plurality of training data sets;
• inputting each of the plurality of new data combinations into the trained plurality of models and evaluating a model accuracy for each of the plurality of models;
• selecting a model from the plurality of models based on the evaluated model accuracy of each of the plurality of models and performing further model training on the selected model, wherein performing the further model training comprises:
• for each training epoch, generating a new random number seed to determine a learning order for the training data sets;
• and
• training the selected model using the training data sets in the determined learning order;
• and
• transmitting a control signal to assign execution of prediction processes using the selected model to the central processing unit responsive to the selected model performing two-class classification and to the graphics processing unit responsive to the selected model performing multi-class classification.

A classification apparatus comprising:

• a processor comprising a central processing unit and a graphics processing unit and configured to:
• receive learning data for model training, wherein the learning data is associated with a plurality of attributes;
• select a target attribute which is an attribute of non-input target data that is to be excluded from model training;
• generate a subset of the learning data, wherein the subset comprises data that is not associated with the target attribute;
• divide the subset of the learning data into a plurality of file sets in chronological order;
• select a subset of file sets from the plurality of file sets in a predetermined order, wherein the selected file sets comprise data that is newer in time series;
• connect the selected file sets based on selection order to generate a training data group;
• generate a plurality of sets of initial values of model parameters by inputting random number seeds into a random function, wherein the random number seeds are selected such that the initial values have a uniform distribution within a predetermined range;
• divide the training data group into a plurality of training data sets according to a buffer size, wherein each training data set comprises fewer data records than a total number of data records in the training data group;
• generate a model having initial values of model parameters from the plurality of sets of initial values;
• for each training epoch, generate a new random number seed to determine a learning order for the training data sets;
• shuffle the plurality of training data sets according to the determined learning order;
• and
• iteratively train the model by sequentially learning features of the plurality of training data sets in the determined learning order;
• and
• transmit a control signal to assign execution of prediction processes using the trained model to the central processing unit responsive to the trained model performing two-class classification and to the graphics processing unit responsive to the trained model performing multi-class classification.