Non-linear model automatic generating method

• US 5,819,246 A
• Filed: 10/17/1995
• Issued: 10/06/1998
• Est. Priority Date: 10/20/1994
• Status: Expired due to Fees

• Current Assignee: Hitachi, Ltd.
• Original Assignee: Hitachi, Ltd.
• Inventors: Tenma, Tadashi, Taniguchi, Yoji, Maeda, Akira, Ashida, Hitoshi
• Primary Examiner(s): MacDonald, Allen R.
• Assistant Examiner(s): Shah, Sanjiv
• Application Number:
  US08/543,998
• Time in Patent Office: 1,085 Days
• Field of Search: 395/22, 395/23, 395/11, 395/900, 706/25
• US Class Current: 706/25
• CPC Class Codes: G06F 18/211 Selection of the most significant subset of features
• CPC Class Codes: G06N 3/105 Shells for specifying net layout

Abstract

The present invention is intended to automatically select least required input items for a non-linear model and improve an efficiency of building up the non-linear model. For building up, for example, a neural network as the non-linear model, a group-by rule 105 and dividing point information 106 of the data are automatically generated by a group-by rule induction device 104 by selecting a dividing method option 101 and a group selection information 103 if data for learning 101 is given. An initial neural network model generating device 107 automatically generates an initial neural network model 108 from the group-by rule 105. The initial neural network model is learned in a neural network model learning device 111 and outputted as a post-learning neural network model 112. Data for learning with group information 110 which is the data for learning is generated in a data classification device 109 by using data for learning 102, group-by rule 105 and dividing point information 106. Input-output variables can be automatically selected from the data for learning according to selection of the group and a neural network model of respective groups can be built up.

Claims

1. A method of automatically generating a non-linear model for generating a non-linear model of a neural network, comprising steps of:

selecting characteristics of learning data to be used for learning a non-linear model as a group-by rule through a group-by rule induction processing;

picking up input-output variables from the selected rule; and

generating a non-linear model of m inputs and an n output (m≧

2, n≧

1) by using the picked-up input-output variables.

2. A method of automatically generating a neural network model for generating a neural network model, comprising steps of:

selecting characteristics of learning data to be used for learning a neural network model as a group-by rule through a group-by rule induction processing;

picking up input-output variables from the selected rule;

generating an initial neural network model for determining a structure of the neural network model by using the picked-up input-output variables; and

generating a neural network model by learning the initial neural network model according to said learning data.

3. A method of automatically generating a neural network model for generating a neural network model, comprising steps of:

inducing characteristics of learning data to be used for learning said neural network model as a group-by rule through a group-by rule induction processing;

generating an initial neural network model for determining a structure of the neural network model by the input-output variables, which are picked up from said group-by rule;

generating learning data with group information by classifying said learning data for respective groups; and

learning said initial neural network model according to the classified learning data.

4. A method of automatically generating a neural network model according to claim 3, wherein said step of inducing comprises steps of:

translating data of numerical attributes in said learning data to an instance table presented with symbols;

selecting a Rule Induction rule from the translated instance table; and

classifying the selected Rule Induction rule into specified groups.

5. A method of automatically generating a neural network model according to claim 4, wherein said step of inducing comprises a step of:

outputting dividing point information denoting said instance table and domains of symbols which represent said instance table.

6. A method of automatically generating a neural network model according to claim 5, wherein said step of outputting the dividing point information comprises a step of:

generating a group-by rule and dividing point information from a dividing method option and said learning data for selecting a method of outputting said dividing point information.

7. A method of automatically generating a neural network model according to claim 5, wherein said step of outputting the dividing point information comprises the steps of:

dividing numerical attribute data in said learning data into a plurality of segmental data by delimiting the data at specified positions for respective attributes; and

allocating specified labels to the respective divided segmental data.

8. A method of automatically generating a neural network model according to claim 7, wherein said step of dividing the numerical attribute data comprises steps of:

displaying a histogram of said data for learning for each input-output variable; and

designating determination of said specified positions by operating a pointer displayed superposed on said histogram.

9. A method of automatically generating a neural network model according to claim 4, wherein said step of classifying the Rule Induction rule for the respective groups comprises a step of:

classifying the Ruel Induction rule for rule, output class or total.

10. A method of automatically generating a neural network model according to claim 3, wherein said step of generating the neural network model comprises steps of:

selecting input-output variables to be used for respective rules for respective groups from said group-by rules; and

generating the initial neural network model from input-output variables of the selected rules.

11. A method of automatically generating a neural network model according to claim 10, wherein said step of generating the initial neural network model comprises steps of:

determining a structure of an existing model from said group-by input-output variables; and

generating the initial neural network model according to the determined structure.

12. A method of automatically generating a neural network model according to claim 3, wherein said step of generating the learning data with group information comprises a step of:

generating the learning data with group information in which the information of a group to which each data belongs is presented to each learning data from said learning data, dividing point information and group-by rule.

13. A method of automatically generating a neural network model according to claim 3, wherein said step of learning the initial neural network model comprises the steps of:

translating said learning data with group information to values included for a specified segment suited for learning of a weight which is a coupling factor between neurons of the neuron network model; and

learning said weight according to a back propagation method by using the translated learning data.

14. A method of automatically generating a neural network model according to claim 13, wherein said step of translating the learning data with group information to said values included in the specified segment comprises a step of:

standardizing numerical data in said learning data with group information to real numbers included in a specified segment.

15. A method of automatically generating a neural network model according to claim 13, wherein said step of learning the weight according to the back propagation method comprises a step of:

learning a weighting factor which is a coupling strength between the neurons of the initial neural network model according to the back propagation method by using the learning data with format post-translation included in a specified segment suited for learning.

16. A method of automatically generating a neural network model according to claim 3, wherein said step of generating the initial neural network model comprises steps of:

selecting the group-by input-output variables from said group-by rules; and

generating the initial neural network model from the selected input-output variables.

17. A method of automatically generating a neural network model according to claim 3, wherein said step of learning the initial neural network model comprises a step of:

generating a post-learning neural network model from said learning data with group information and said initial neural network model.