System and method for performing compound computational experiments

• US 6,961,688 B2
• Filed: 03/12/2002
• Issued: 11/01/2005
• Est. Priority Date: 09/16/1997
• Status: Expired due to Fees

• Current Assignee: Evolving Logic Associates
• Original Assignee: Evolving Logic Associates
• Inventors: Bankes, Steven C.
• Primary Examiner(s): Broda, Samuel
• Application Number:
  US10/097,087
• Publication Number:
  US 20020147577A1
• Time in Patent Office: 1,330 Days
• Field of Search: 703 1- 2, 703/6, 703 13- 22, 706/45, 707/103.R
• US Class Current: 703/6
• CPC Class Codes: G06N 3/126 Evolutionary algorithms, e.g. genetic algorithms or genetic programming
• CPC Class Codes: Y10S 707/99944 Object-oriented database structure

Abstract

A system and method for performing compound computational experiments include an experiment having at least one input. The experiment includes a process controllable by a computer and responsive to the input to produce at least one output, the output capable of being stored on a medium readable by the computer. The system includes a generator capable of running on the computer. The generator automates the performance of a plurality of experiments and automatically varies the inputs for the experiments. An outcome store of the system stores the outputs on the computer-readable medium. The system also includes a display capable of running on the computer. The display presents an interpretation of a set of the outputs extracted from the outcome store. The display is responsive to user input to alter the interpretation, to present an interpretation of a different set of outputs, or to cause the generator to perform a plurality of experiments.

Claims

1. A method for performing compound computational experiments, comprising the steps of:

encapsulating an experiment in an experiment object having a run method, said run method callable by other objects to run an experiment, said run method encapsulating outputs produced by an experiment in a result object;

encapsulating all input and output variables for an experiment in a domain object, said domain object having a read method callable by other objects to receive information about the input and output variables;

encapsulating a first search or sampling strategy in a first generator object, said first generator object having a generate method callable by other objects to repetitively encapsulate a value for each input variable in a specification object and to repetitively call said run method to perform an experiment using each specification object, said generator calling said read method to acquire information about the input variables;

encapsulating experiment result data in an outcomeStore object, said outcomeStore object having an extract method callable by other objects to retrieve experiment result data; and

requesting a plot of experiment result data by issuing a plot command through a user interface running on a computer, said request invoking said generate method to produce experiment result data and invoking said extract method to acquire experiment result data for experiment dimensions related to said requested plot.

2. The method described in claim 1, comprising the further steps of:

plotting pixel data on a computer screen to visualize experiment result data corresponding to said experiment dimensions related to said requested plot; and

adjusting a control on said user interface to associate a value with an unplotted dimension of said experiment, said adjustment invoking said generate method to produce second experiment result data, invoking said extract method to acquire second experiment result data for experiment dimensions related to said requested plot, and plotting pixel data on said computer screen to visualize said second experiment result data corresponding to said experiment dimensions related to said requested plot.

3. The method described in claim 1, wherein encapsulating a first search or sampling strategy comprises providing a representation of at least one of Monte Carlo sampling, factorial sampling, Latin Hypercube sampling, stochastic hill climbing, gradient descent, or a genetic algorithm.

4. The method described in claim 1, further comprising:

encapsulating a second search or sampling strategy in a second generator object; and

performing the encapsulated experiment using the first generator object and performing the encapsulated experiment using the second generator object such that results of the encapsulated experiment using the first generator object and results of the encapsulated experiment using the second generator object can be compared.

5. The method described in claim 4, further comprising allowing a user to choose the first search or sampling strategy and the second search or sampling strategy.

6. The method described in 4, further comprising performing at least one transformation operation on the outcomeStore object.

7. The method of claim 6, wherein performing at least one transformation operation on the outcomeStore object results in a data object that has fewer dimensions than the outcomeStore object.

8. A method for performing compound computational experiments, the method comprising:

providing a plurality of data structures configured to;

encapsulate an experiment in an experiment object having a run method, said run method callable by other objects to run an experiment, said run method encapsulating outputs produced by an experiment in a result object;

encapsulate all input and output variables for an experiment in a domain object, said domain object having a read method callable by other objects to receive information about the input and output variables;

encapsulate a plurality of search or sampling strategies in a corresponding plurality of generator objects, each of said generator objects having a generate method callable by other objects to repetitively encapsulate a value for each input variable in a specification object and to repetitively call said run method to perform an experiment using each specification object, said generator calling said read method to acquire information about the input variables; and

encapsulate experiment result data in an outcomeStore object, said outcomeStore object having an extract method callable by other objects to retrieve experiment result data;

selecting at least one of the encapsulated search or sampling strategies and performing a compound computational experiment using a generator object corresponding to the selected encapsulated search or sampling strategy; and

requesting a plot of experiment result data.

9. The method of claim 8, wherein the selected at least one encapsulated search or sampling strategy comprises a representation of at least one of Monte Carlo sampling, factorial sampling, Latin Hypercube sampling, stochastic hill climbing, gradient descent, or a genetic algorithm.

10. The method of claim 8, wherein selecting at least one of the encapsulated search or sampling strategies comprises receiving a user selection of at least one of the encapsulated search or sampling strategies.

11. The method of claim 8, wherein the selected at least one encapsulated search or sampling strategy is a first encapsulated search or sampling strategy, further comprising selecting a second encapsulated search or sampling strategy and performing the compound computational experiment using a generator object corresponding to the second selected encapsulated search or sampling strategy.

12. The method of claim 8, further comprising performing at least one transformation operation on the outcomeStore object.

13. The method of claim 12, wherein performing at least one transformation operation on the outcomeStore object results in a data object that has fewer dimensions than the outcomeStore object.