Facilitated reuse of K locations in a knowledge store

• US 7,409,380 B1
• Filed: 04/07/2005
• Issued: 08/05/2008
• Est. Priority Date: 04/07/2005
• Status: Expired due to Fees

• Current Assignee: Unisys Corporation
• Original Assignee: Unisys Corporation
• Inventors: Claar, Jane Van Keuren, Mazzagatti, Jane Campbell, Rajcan, Steven L., Collins, Timothy William
• Primary Examiner(s): MAHMOUDI, HASSAN
• Application Number:
  US11/100,830
• Time in Patent Office: 1,216 Days
• Field of Search: 707/1, 707/100
• US Class Current: 1/1
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: Y10S 707/99931 Database or file accessing

Abstract

In learning for an interlocking trees datastore or KStore, the process is made more efficient by noting the (n-level) address within the KStore during the learning of each particle. In a pre-particle stream of data, which may be organized within or before the Learn Engine prior to this, “Marks” and “References” are inserted. Each Mark identifies where any number of References may start the learning process, enabling the avoidance of re-learning redundant data. Thus, in a field record data set, the redundant data fields (or even partial fields) can be skipped over and only the new data learned. The Marks and References are removed before processing into a particle stream. When particles are learned the K Engine returns the n-level address or pointer(s) which the Learn Engine uses to associate with the relevant Reference(s). The system can be implemented in hardware if desired to speed processing. No limit to the distribution or numbers of KStores, Learn Engines being used or K Engines being used is indicated.

Claims

1. An improved method of operating a learn unction which provides particlized messages to a K Engine for learning by an interlocking trees datastore, wherein said learn function receives data input streams in sequences, and provides serial particle streams of said particlized messages to said K Engine and recording said data input streams into said interlocking trees datastore, said method comprising;

preparing a serial pre-particlization sequence incorporating control characters for each desired Mark and Reference in said serial pre-particlization sequence,

preparing a learn function accessible memory structure identifying at least one K location to each Mark,

finding Marks and References in said prepared serial pre-particlization sequence,

processing said found Marks and References,

preparing particles and recording said prepared particles in the interlocking trees datastore from said pre-particlization sequence, incorporating a K location for each Reference, the interlocking trees datastore comprising nodes and non-root nodes organized into a plurality of connected trees, the plurality of connected trees comprising an asCase tree and at least one of a plurality of asResult trees, the asCase tree comprising a first root and at least one of a plurality of non-root nodes, the at least one non-root node comprising a first Case node, and the at least one asResult tree comprising a second root node representing a particle of the prepared particles and at least one of a plurality of non-root nodes common with the nodes of the asCase tree, the at least one non-root node common with the nodes of the asCase tree comprising a first Result node.

2. The method of claim 1 wherein each said sequence of an input data stream comprises a record from a field record database.

3. The method of claim 1 wherein said preparing a learn function accessible memory structure comprises configuring an array of said K locations identified to said each Mark and using said array to produce a particular K location for each said Reference.

4. The method of claim 1 further comprising sending said particles to said K Engine and receiving back by said learn function from said K Engine a K location for each said particle so sent.

5. The method of claim 4 wherein said processing of said found Marks and References further comprises;

recording in said learn function accessible memory structure the K location returned from a K Engine, wherein said K Engine has recorded a particle originating from just prior to a Mark in said pre-particlization stream, and

if a Reference associated with said Mark appears in a pre-particlization sequence, employing said returned K location to establish an address in K for the particle originating from data which followed the Reference in the pre-particlization sequence.

6. The method of claim 1 wherein said K location comprises a pointer having an address for a node at each level in said interlocking trees datastore.

7. The method of claim 1 wherein said K location comprises a pointer to at least one node from one level in said interlocking trees datastore.

8. The method of claim 1 wherein an application level component creates said input data streams for said learn function from at least one data source and in so doing associates Marks and References to data in said input data streams from said at least one data source.

9. The method of claim 8 wherein said application level component receives data from said data source in data source sending format and configures it to be in a format that has only the data desired.

10. The method of claim 1 wherein a component internal to the learn function associates Marks and References to data in said input data streams from said at least one data source.

11. The method of claim 10 wherein said component internal to said learn function receives said input data streams from said at least one data source in data source sending format and configures it to be in a format having only data desired to be learned by a designer of a particular interlocking trees datastore, and in a predetermined organizational format before associating Marks and References to data in said input data streams.

12. The method of claim 1 further comprising finding metadata in said input data stream in said learn function, and based on said metadata, inserting Marks and References into said input data stream at locations indicated by said metadata.

13. The method of claim 12 further comprising preconfiguring said learn function to adapt to said metadata.

14. The method of claim 1 further comprising preconfiguring said Learn Engine to interpret an input data stream so as to predetermine insertion points for Marks and References based on preset understanding of a formatting of said input data stream.

15. The method of claim 1 wherein when an input data stream includes tabular data, at least each new row or column of said tabular data is associated to a single Mark by a new Reference.

23. A computer readable memory having instructions for configuring a computer system when activated on said computer system a process as set forth in claim 1.

16. A computer facilitated learning function system component for providing particlized messages to a K Engine, which K Engine is providing learning data input from said learning function system component to an interlocking trees datastore, said learning function system component comprising;

first process control component for inserting Marks and References into an input data stream in accord with predetermined controls for location of said Marks and References within said input data stream, said first process control component producing an ordered pre-particlized sequence of data from said input data stream having said Marks and References located therein in accord with said predetermined controls,

second process control component for removing Marks and References from said pre-particlized sequence of data while building particle messages from said pre-particlized sequence, and for each Reference removed, inserting a K location into a message wherein said K location is related to said removed Reference via a Mark related to said removed Reference, and

a storage component located in a memory for storing said Marks to enable association of each such Mark to a Reference and recording said ordered pre-particlized sequence of data in the interlocking trees datastore, the interlocking trees datastore comprising nodes and non-root nodes organized into a plurality of connected trees, the plurality of connected trees comprising an asCase tree and at least one of a plurality of asResult trees, the asCase tree comprising a first root and at least one of a plurality of non-root nodes, the at least one non-root node comprising a first Case node, and the at least one asResult tree comprising a second root nod representing a particle of the pre-particlized sequence of data and at least one of a plurality of non-root nodes common with the nodes of the asCase tree, the at least one non-root node common with the nodes of the asCase tree comprising a first Result node.

17. The learning function system component of claim 16 wherein said first process control component is for inserting a particular kind of control character for each Mark, each such control character for a Mark having an associated index value through which each of said References to a related one of said Marks, and wherein said first process control component also is for inserting a second particular kind of control character for each Reference, each Reference having an associated index value through which it may be related to a particular Mark.

18. The learning function system component of claim 16 further comprising an interpretive component for determining from an input data stream where to cause said first process control component to insert said Marks and References.

19. The learning function system component of claim 16 further comprising a configurator component for formatting input data streams into predetermined order before providing said input data streams into said first process control component.

20. The learning function system component of claim 16 further comprising an access through a GUI for determining functionality of components of said learning function system component, said components of said learning function system component including but not limited to a configurator component, a data stripper component, an interpretive component, said first process control component, and said second process component.

21. A computer readable storage medium including digitally encoded data readable by a processing system having a KStore interlocking trees datastore, stored therein, a learning function system component as set forth in claim 16, and a K Engine, for communicating between said KStore interlocking trees datastore and said learning function system component.

22. The computer readable storage medium of claim 21 having instructions for configuring a computer system when activated on said computer system a computer system as set forth in claim 21.