Classification and non-parametric regression framework with reduction of trained models

Patent metadata

Field	Value
Publication number	US-9501749-B1
Application number	US-201313799517-A
Country	US
Kind code	B1
Filing date	Mar 13, 2013
Priority date	Mar 14, 2012
Publication date	Nov 22, 2016
Grant date	Nov 22, 2016

Abstract

A device receives selection of a classification and regression framework, and receives training data for the classification and regression framework. The device applies the training data to the classification and regression framework to generate a trained model, and monitors performance of the trained model. The device inspects a structure of the trained model, and reduces a size of the trained model. The device generates an object based on the trained model, and provides the object for display.

First claim

What is claimed is:

1. A method comprising:
interacting with a selection of a classification and regression framework,
the classification and regression framework providing a plurality of learning algorithms, and
the interacting with the selection being performed by a device;

applying training data to the classification and regression framework to generate a first trained model,
the training data including information to discover predictive relationships,
the first trained model including a plurality of trained learning algorithms, and
the applying the training data being performed by the device;

reducing, based on weights associated with the plurality of trained learning algorithms, a size of the first trained model to form a second trained model including fewer trained learning algorithms than the first trained model,
the reducing the size of the first trained model including removing at least one of the plurality of trained learning algorithms from the first trained model, and
the reducing the size of the first trained model being performed by the device;

generating a compact object, for use in producing predicted results and based on the second trained model, that does not include the training data,
the generating the compact object being performed by the device; and

providing the compact object for display,
the providing the compact object for display being performed by the device.

2. The method of claim 1 , further comprising:
storing the compact object.

3. The method of claim 1 , further comprising:
applying the compact object to data;
generating, via the compact object, the predicted results for the data; and
providing the predicted results for display.

4. The method of claim 1 , further comprising:
receiving a request to add a new model to the classification and regression framework;
creating, based on the request, a first class that determines predictions of the new model based on particular data;
creating a second class that stores input parameters for the new model;
creating a third class that builds the new model based on training data and class labels; and
adding the new model, the first class, the second class, and the third class to a set of models of the classification and regression framework.

5. The method of claim 4 , further comprising:
creating a standalone model, for the new model, based on the classification and regression framework,
the standalone model providing functionality without the classification and

regression framework; and
storing the standalone model.

6. The method of claim 1 , where the classification and regression framework is associated with a technical computing environment.

7. A non-transitory computer-readable medium storing instructions, the instructions comprising:
one or more instructions that, when executed by a processor of a device, cause the processor to:
interact with a selection of a classification and regression framework,
the classification and regression framework providing a plurality of learning algorithms,

apply training data to the classification and regression framework to generate a first trained model,
the training data including information to discover predictive relationships, and
the first trained model including a plurality of trained learning algorithms,

reduce, based on weights associated with the plurality of trained learning algorithms, a size of the first trained model to form a second trained model including fewer learning algorithms than the first trained model,
the reducing the size of the first trained model including removing at least one of the plurality of trained learning algorithms from the first trained model,

generate a compact object, for producing predicted results and based on the second trained model, that does not include the training data, and
provide the compact object for display.

8. The non-transitory computer-readable medium of claim 7 , where the instructions further comprise:
one or more instructions that, when executed by the processor, cause the processor to:
store the compact object.

9. The non-transitory computer-readable medium of claim 7 , where the instructions further comprise:
one or more instructions that, when executed by the processor, cause the processor to:
apply the compact object to data,
generate, via the compact object, the predicted results for the compact object, and
provide the predicted results for display.

10. The non-transitory computer-readable medium of claim 7 , where the instructions further comprise:
one or more instructions that, when executed by the processor, cause the processor to:
receive a request to add a new model to the classification and regression framework,
create, based on the request, a first class that determines predictions of the new model based on particular data,
create a second class that stores input parameters for the new model,
create a third class that builds the new model based on training data and class labels, and
add the new model, the first class, the second class, and the third class to a set of models of the classification and regression framework.

11. The non-transitory computer-readable medium of claim 10 , where the instructions further comprise:
one or more instructions that, when executed by the processor, cause the processor to:
create a standalone model, for the new model, based on the classification and regression framework,
the standalone model providing functionality without the classification and regression framework, and

store the standalone model.

12. The non-transitory computer-readable medium of claim 11 , where the instructions further comprise:
one or more instructions that, when executed by the processor, cause the processor to:
provide the standalone model for display.

13. The non-transitory computer-readable medium of claim 7 , where the classification and regression framework is associated with a technical computing environment.

14. A device comprising:
one or more processors to:
interact with a selection of a classification and regression framework,
the classification and regression framework providing a plurality of learning algorithms,

apply training data to the classification and regression framework to generate a first trained model,
the training data including information to discover predictive relationships, and
the first trained model including a plurality of trained learning algorithms,

reduce, based on weights associated with the plurality of trained learning algorithms, a size of the first trained model to generate a second trained model including fewer trained learning algorithms than the first trained model,
the reducing the size of the first trained model including removing at least one of the plurality of trained learning algorithms from the first trained model,

generate a compact object, for producing predicted results and based on the second trained model, that does not include the training data, and
provide the compact object for display.

15. The device of claim 14 , where the one or more processors are further to:
store the compact object.

16. The device of claim 14 , where the one or more processors are further to:
apply the compact object to data,
generate, via the compact object, the predicted results for the data, and
provide the predicted results for display.

17. The device of claim 14 , where the one or more processors are further to:
receive a request to add a new model to the classification and regression framework,
create, based on the

Assignees

• Mathworks Inc

Inventors

• Narsky Ilya

Classifications

• G06N99/005 Primary

  Physics · mapped topic

• G06N7/023

  Learning or tuning the parameters of a fuzzy system · CPC title

• G06N20/20 Primary

  Ensemble learning · CPC title

• G06F8/20

  Software design · CPC title

• G06N20/00 Primary

  Machine learning · CPC title