Distributed evolutionary algorithm for asset management and trading

• US 8,527,433 B2
• Filed: 04/28/2010
• Issued: 09/03/2013
• Est. Priority Date: 04/28/2009
• Status: Active Grant

• Current Assignee: Cognizant Technology Solutions U.S. Corporation (Cognizant Technology Solutions Corp.)
• Original Assignee: Genetic Finance (Barbados) Limited
• Inventors: Hodjat, Babak, Shahrzad, Hormoz
• Primary Examiner(s): CHEN, ALAN S
• Application Number:
  US12/769,589
• Publication Number:
  US 20100274742A1
• Time in Patent Office: 1,224 Days
• Field of Search: None
• US Class Current: 706/12
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 3/086 using evolutionary algorithms, e.g. genetic algorithms or genetic programming
• CPC Class Codes: G06N 3/12 using genetic models
• CPC Class Codes: G06N 3/126 Evolutionary algorithms, e.g. genetic algorithms or genetic programming
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising

Abstract

A server computer and a multitude of client computers form a network computing system that is scalable and adapted to continue to evaluate the performance characteristics of a number of genes generated using a software application running on the client computers. Each client computer continues to periodically receive data associated with the genes stored in its memory. Using this data, the client computers evaluate the performance characteristic of their genes by comparing a solution provided by the gene with the periodically received data associated with that gene. Accordingly, the performance characteristic of each gene may be updated and varied with each periodically received data. The performance characteristic of a gene defines its fitness. The genes may be virtual asset traders that recommend trading options, and the data associated with the genes may be historical trading data.

Claims

1. A networked computer system comprising a first client computer, the first client computer comprising:

a memory operative to store N genes, each gene characterized by a plurality of conditions and at least one action, wherein N is an integer greater than one;

a communication port through which the first client computer recurrently receives data associated with the N genes; and

a processor operative to evaluate performance characteristic of each of the N genes by comparing a solution provided by each gene with the recurrently received data associated with that gene, the performance characteristic of each gene being adjusted with each recurrently received data and defining a fitness of the gene,

wherein the data associated with the N genes comprise historical trading data and wherein the solution provided by each gene comprises a trading recommendation made by the gene,

wherein said first client computer is configured to discard M of the N genes after evaluating the fitness of the N genes for P trading days, each of the discarded genes having a fitness that falls below a first predefined threshold value, each of the remaining N-M genes being a surviving gene, wherein M and P are positive integers and wherein M is smaller than N,

and wherein said first client computer is further configured to;

evaluate, for a plurality of Q trading days in addition to the P trading days, the fitness of the remaining genes;

discard genes whose fitness as evaluated during a sum of the P+Q trading days is below a second predefined threshold value; and

repeat the evaluation and the discard operations S times, wherein Q and S are integers, and wherein S is equal to or greater than 1, wherein genes surviving S evaluation and discard operations define a first elitist gene pool; and

transmit the first elitist pool of genes to a server computer.

2. The networked computer system of claim 1 wherein the first and second predefined threshold values are equal.

3. The networked computer system of claim 1, comprising a plurality of client computers including said first client computer, and further comprising said server computer, wherein said server computer is adapted to:

receive the first elitist gene pool from the first client computer;

store the received genes in a server gene pool; and

transmit to each of at least a first subset of the plurality of client computers one or more genes of the server gene pool for fitness evaluation spanning W trading days.

4. The networked computer system of claim 3 wherein each of the at least a first subset of the plurality of client computers is adapted to:

receive the one or more genes of the server gene pool from the server computer;

evaluate a fitness of the received genes by comparing trading recommendations made by each of the received genes with an associated historical trading data spanning the W trading days;

discard genes whose fitness are less than a third threshold value; and

transmit genes that are not discarded to the server computer for consideration for storage in the server gene pool.

5. The networked computer system of claim 4 wherein said W trading days do not overlap the trading days on which the genes in the first elitist gene pool were evaluated.

6. The networked computer system of claim 4 wherein said third threshold value is equal to the first and second threshold values.

7. The networked computer system of claim 4 wherein said genes transmitted to the server computer which were not discarded exclude all genes whose fitness is no-higher than the fitness of a least fit gene previously stored in the server gene pool.

8. The networked computer system of claim 7 wherein said server computer is further adapted to combine a plurality of fitness values associated with each gene the server computer receives from the at least a first subset of the plurality of client computers with a corresponding fitness value stored in the server computer for that gene.

9. The networked computer system of claim 4 wherein said server gene pool is configured to store a fixed number of genes.

10. The networked computer system of claim 1 wherein the N genes stored in the memory of the client computer are generated in accordance with computer instructions stored in the first client computer and executed by the processor of the first client computer.

11. A networked computer system comprising a server computer, said server computer configured to:

receive a first gene pool from a first one of a plurality of client computers;

store received genes in a server gene pool; and

transmit to each of at least a first subset of the plurality of client computers one or more genes of the server gene pool for fitness evaluation spanning W trading days, the first gene pool being generated by evaluating performance characteristic of each of N genes by comparing a solution provided by each gene with the recurrently received data associated with the gene, the performance characteristic of each gene being adjusted with each recurrently received data and defining a fitness of the gene, each gene identifying by a plurality of conditions and at least one action,

wherein the data associated with each gene comprises historical trading data and wherein the solution provided by each gene comprises a trading recommendation made by the gene,

wherein said first one of the plurality of client computers is configured to discard M of the N genes after evaluating the fitness of the N genes for P trading days, each of the discarded genes having a fitness that falls below a first predefined threshold value, each of the remaining N-M genes being a surviving gene, wherein M and P are positive integers and wherein M is smaller than N,

and wherein said first one of the plurality of client computers is further configured to;

evaluate, for a plurality of Q trading days in addition to the P trading days, the fitness of the remaining genes;

discard the genes whose fitness as evaluated during a sum of the P+Q trading days is below a second predefined threshold value; and

repeat the evaluation and the discard operations S times, wherein Q and S are integers, and wherein S is equal to or greater than 1, wherein genes surviving S evaluation and discard operations define the first gene pool.

12. The networked computer system of claim 11 wherein said server computer accepts only genes whose fitness is higher than the fitness of a least fit gene previously stored in the server gene pool.

13. The networked computer system of claim 12 wherein said server gene pool is configured to store a fixed number of genes.

14. The networked computer system of claim 12 wherein said server computer is further adapted to combine a plurality of fitness values associated with each gene the server computer receives with a corresponding fitness value stored in the server computer for that gene.

15. The networked computer system of claim 11, further comprising said first client computer.

16. A method for solving a computational problem using a plurality of client computers, the method comprising:

storing N genes in a memory accessible to a first one of said client computers, each gene characterized by a plurality of conditions and at least one action, wherein N is an integer greater than one;

through a communications port, recurrently receiving data associated with the N genes;

said first client computer evaluating performance characteristic of each of the N genes by comparing a solution provided by each gene with the recurrently received data associated with that gene, the performance characteristic of each gene being adjusted with each recurrently received data and defining a fitness of the gene, wherein the data associated with each gene comprises historical trading data and wherein the solution provided by each gene comprises a trading recommendation made by the gene;

discarding M of the N genes after evaluating the fitness of the N genes for P trading days, each of the discarded genes having a fitness that falls below a first predefined threshold value, each of the remaining N-M genes being a surviving gene, wherein M and P are positive integers and wherein M is smaller than N;

evaluating, for a plurality of Q trading days in addition to the P trading days, the fitness of the remaining genes;

discarding genes whose fitness as evaluated during a sum of the P+Q trading days is below a second predefined threshold value;

repeating the evaluation and the discard operations S times, wherein Q and S are integers, and wherein S is equal to or greater than 1, wherein genes surviving S evaluation and discard operations define a first elitist gene pool; and

transmitting the first elitist pool of genes to a server computer.

17. The method of claim 16 wherein the first and second predefined threshold values are equal.

18. The method of claim 16 further comprising:

receiving the first elitist gene pool from the first one of the plurality of client computer;

storing the received genes in a server gene pool; and

transmitting to each of at least a first subset of the plurality of client computers one or more genes of the server gene pool for fitness evaluation spanning W trading days.

19. The method of claim 18 wherein said W trading days and V trading days do not overlap.

20. The method of claim 18 wherein said third threshold value is equal to the first and second threshold values.

21. The method of claim 18 wherein said genes transmitted to the server computer which were not discarded exclude all genes whose fitness is no higher than the fitness of a least fit gene previously stored in the server gene pool.

22. The method of claim 21 further comprising:

combining a plurality of fitness values associated with each gene the server computer receives from the at least a first subset of the plurality of client computers with a corresponding fitness value stored in the server computer for that gene.

23. The method of claim 18 wherein said server pool is configured to store a fixed number of genes.

24. The method of claim 16 wherein the N are generated in accordance with computer instructions stored in the first client computer and executed by the first client computer.

25. A stand alone computer comprising:

a processing core configured as a server;

and a plurality of processing cores configured as clients, each of the plurality of clients comprising;

a memory operative to store N genes, each gene characterized by a plurality of conditions and at least one action, wherein N is an integer greater than one, and

a port through which the first client processing core recurrently receives data associated with the N genes,

each client processing core configured to evaluate performance characteristic of each of the N genes by comparing a solution provided by each gene with the recurrently received data associated with that gene, the performance characteristic of each gene being adjusted with each recurrently received data and defining a fitness of the gene, wherein the data associated with each gene comprises historical trading data and wherein the solution provided by each gene comprises a trading recommendation made by the gene,

wherein a first one of the plurality of client computers is configured to discard M of the N genes after evaluating the fitness of the N genes for P trading days, each of the discarded genes having a fitness that falls below a first predefined threshold value, each of the remaining N-M genes being a surviving gene, wherein M and P are positive integers and wherein M is smaller than N,

wherein said first one of the client computers is further configured to;

evaluate, for a plurality of Q trading days in addition to the P trading days, the fitness of the remaining genes;

discard genes whose fitness as evaluated during a sum of the P+Q trading days is below a second predefined threshold value; and

repeat the evaluation and the discard operations S times, wherein Q and S are integers, and wherein S is equal to or greater than 1, wherein genes surviving S evaluation and discard operations define a first elitist gene pool; and

transmit the first elitist pool of genes to the server processing core.

26. The stand alone computer of claim 25 wherein the first and second predefined threshold values are equal.

27. The stand alone computer of claim 25 wherein said server processing core is adapted to:

receive the first elitist gene pool from the first client processing core;

store the received genes in a server processing core gene pool; and

transmit to each of at least a first subset of the plurality of client processing cores one or more genes of the server processing core gene pool for fitness evaluation spanning W trading days.

28. The stand alone computer of claim 25 wherein each of the at least a first subset of the plurality of client processing cores is adapted to:

receive the one or more genes of the first elitist gene pool from the server processing core;

evaluate a fitness of the received genes by comparing trading recommendations made by each of the received genes with an associated historical trading data spanning the W trading days;

discard genes whose fitness are less than a third threshold value; and

transmit genes that are not discarded to the server processing core for consideration for storage in a server processing core gene pool.

29. The stand alone computer of claim 28 wherein said W trading days do not overlap the trading days on which the genes in the first elitist gene pool were evaluated.

30. The stand alone computer of claim 28 wherein said third threshold value is equal to the first and second threshold values.

31. The stand alone computer of claim 28 wherein said genes transmitted to the server processing core which were not discarded exclude all genes whose fitness is no higher than the fitness of a least fit gene previously stored in the server processing core gene pool.

32. The stand alone computer of claim 31 wherein said server processing core is further adapted to combine a plurality of fitness values associated with each gene the server processing core receives from the at least first subset of the plurality of client processing cores with a corresponding fitness value stored in the server processing core for that gene.

33. The stand alone computer of claim 28 wherein said server processing core gene pool is configured to store a fixed number of genes.

34. The stand alone computer of claim 25 wherein the N genes stored in each client processing core are generated in accordance with computer instructions stored in and executed by the stand alone computer.

35. The stand alone computer of claim 25 wherein said plurality of conditions are evaluated as a logical expression conjoined by logical AND and/or modified by logical NOT operations.

36. The stand alone computer of claim 35 wherein the at least one action of each gene is selected from a group consisting of buy, sell, hold, long exit and short exit recommendations.

37. The stand alone computer of claim 25 wherein said stand alone computer is a mainframe computer.