Computation platform agnostic data classification workflows

• US 10,698,954 B2
• Filed: 06/30/2016
• Issued: 06/30/2020
• Est. Priority Date: 06/30/2016
• Status: Active Grant

• Current Assignee: Meta Platforms, Inc. (f/k/a Facebook, Inc.)
• Original Assignee: Meta Platforms, Inc. (f/k/a Facebook, Inc.)
• Inventors: Piechowicz, Szymon, Naveh, Barak Reuven, Liu, Annie Hsin-Wen, Gupta, Ashish
• Primary Examiner(s): Trujillo, James
• Assistant Examiner(s): Le, Jessica N
• Application Number:
  US15/199,351
• Publication Number:
  US 20180004859A1
• Time in Patent Office: 1,461 Days
• Field of Search: 707798
• CPC Class Codes: G06F 16/9024 Graphs; Linked lists G06F16/9027 takes precedence

Abstract

Various embodiments include a classification platform system. A user can define a classification experiment on the classification platform system. For example, the user can define an input data space by selecting at least one of data sources interfaced with the classification platform system and defining a workflow configuration including a directed graph (DG) connecting a plurality of transformation blocks to represent an experiment workflow. The DG can specify how one or more outputs of each of the transformation blocks are fed into one or more other transformation blocks. The DG can be executed by various types of computation platforms. The classification platform system can schedule the experiment workflow to be executed on a distributed computation platform according to the input data space and the workflow configuration.

Claims

1. A computer-implemented method, comprising:

defining a classification experiment for executing tasks on a computation platform by at least;

defining an input data space by selecting at least one of data sources interfaced with a classification platform system; and

defining, via a definition user interface of the classification platform system, a workflow configuration of the classification experiment by defining a directed graph (DG) connecting a plurality of transformation blocks to represent an experiment workflow, wherein the DG specifies one or more connections from one or more outputs of each transformation block in the experiment workflow to one or more other transformation blocks;

formatting the workflow configuration and the input data space into a data structure such that the data structure is interpretable by a plurality of different computation platforms to execute the classification experiment, wherein the plurality of different computation platforms have a plurality of different hardware configurations of system components, respectively;

selecting, from the plurality of different computation platforms, a distributed computation platform to execute at least part of the experiment workflow in accordance with the DG;

generating, based on a hardware configuration of system components associated with the distributed computation platform, an execution schedule that specifies which system component executes each part of the classification experiment; and

scheduling the distributed computation platform to execute the classification experiment by the system components under the execution schedule according to the input data space and the workflow configuration, wherein the system components are scheduled to execute their respective parts of the classification experiment as specified by the execution schedule.

2. The computer-implemented method of claim 1, wherein the DG is arranged graphically via the definition user interface and wherein the definition user interface is a graphical user interface.

3. The computer-implemented method of claim 1, wherein said scheduling includes scheduling a first part of the workflow configuration to execute on a first computation platform and a second part of the workflow configuration to execute on a second computation platform.

4. The computer-implemented method of claim 1, wherein the distributed computation platform is selected based on a geographical or network location of the distributed computation platform relative to one or more geographical or network locations of input data specified in the input data space.

5. The computer-implemented method of claim 1, further comprising maintaining a memorization database;

wherein said scheduling includes preventing a transformation block from being executed by the distributed computation platform when the transformation block as defined by the workflow configuration matches an entry in the memorization database; and

wherein the entry includes pre-computed output result of the transformation block given the same input and configuration.

6. The computer-implemented method of claim 1, wherein the data sources include a live data source from a social networking system, and wherein the live data source produces an open-ended stream of new data entries formatted according to one or more data formats of the defined input data space.

7. The computer-implemented method of claim 1, wherein the data sources include a static data source from a social networking system, and wherein the static data source includes a static data set with a constant data size formatted according to one or more data formats of the defined input data space.

8. The computer-implemented method of claim 1, wherein the DG is acyclical and thereby prevents execution of the classification experiment to enter an infinite loop.

9. The computer-implemented method of claim 1, wherein a transformation block in the DG includes logic to dynamically modify the DG during execution of the experiment workflow.

10. The computer-implemented method of claim 9, wherein the transformation block includes logic to dynamically modify input data of an existing transformation block in the DG.

11. The computer-implemented method of claim 9, wherein the transformation block includes logic to change or remove an existing transformation block in the DG or to add a new transformation block to the DG.

12. The computer-implemented method of claim 1, further comprising piping an output result of executing the classification experiment to a social networking system to re-configure at least an application service of the social network system.

13. The computer-implemented method of claim 1, wherein the input data space is a labeled data space that includes at least a parameter to locate labeled data for training a supervised classifier machine learning model or for evaluating classification precision or recall of a classifier model, wherein said training or said evaluating is represented in a transformation block in the DG.

14. The computer-implemented method of claim 1, wherein the input data space is a prediction space that includes at least a parameter to locate input data to be classified in the classification experiment.

15. The computer-implemented method of claim 1, wherein defining the classification experiment further includes defining a domain configuration that includes at least a parameter binding the input data space to the workflow configuration.

16. The computer-implemented method of claim 1, wherein defining the classification experiment includes inheriting a directed graph for the workflow configuration from a workflow repository.

17. The computer-implemented method of claim 1, wherein the input data space selects at least a live data source from the data sources to feed into at least one of the transformation blocks;

and wherein the distributed computation platform is configured to execute the classification experiment in substantially real-time in response to new data from the live data source.

18. The computer-implemented method of claim 1, wherein the transformation blocks include a data feature extraction process, data feature filtering process, a data feature transformation process, a classifier deliberation process, a classifier training process, a classifier evaluation process, or any combination thereof.

19. The method of claim 1, wherein formatting the workflow configuration is based on the DG.

20. A computer readable non-transitory data storage memory storing computer-executable instructions that, when executed by a computer system, cause the computer system to perform a computer-implemented method, the computer-executable instructions comprising:

instructions for defining a classification experiment for executing tasks on a computation platform by at least;

defining an input data space by selecting at least one of data sources interfaced with a classification platform system; and

defining, via a definition user interface of the classification platform system, a workflow configuration of the classification experiment by defining a directed graph (DG) connecting a plurality of transformation blocks to represent an experiment workflow, wherein the DG specifies one or more connections from one or more outputs of each transformation block in the experiment workflow to one or more other transformation blocks;

instructions for formatting the workflow configuration and the input data space into a data structure such that the data structure is interpretable by a plurality of different computation platforms to execute the classification experiment, wherein the plurality of different computation platforms have a plurality of different hardware configurations of system components, respectively;

instructions for selecting, from the plurality of different computation platforms, a distributed computation platform to execute at least part of the experiment workflow in accordance with the DG;

instructions for generating, based on a hardware configuration of system components associated with the distributed computation platform, an execution schedule that specifies which system component executes each part of the classification experiment; and

instructions for scheduling the distributed computation platform to execute the classification experiment by the system components under the execution schedule according to the input data space and the workflow configuration, wherein the system components are scheduled to execute their respective parts of the classification experiment as specified by the execution schedule.

21. A classification platform system, comprising:

a memory configured to store executable instructions; and

a processor configured by the executable instructions to implement an experiment management engine and a workflow execution engine;

wherein the experiment management engine is configured to defining a classification experiment for executing tasks on a computation platform by at least;

defining an input data space by selecting at least one of data sources interfaced with a classification platform system; and

defining, via a definition user interface of the classification platform system, a workflow configuration of the classification experiment by defining a directed graph (DG) connecting a plurality of transformation blocks to represent an experiment workflow, wherein the DG specifies one or more connections from one or more outputs of each transformation block in the experiment workflow to one or more other transformation blocks;

wherein the workflow execution engine is configured to format the workflow configuration and the input data space into a data structure such that the data structure is interpretable by a plurality of different computation platforms to execute the classification experiment, wherein the plurality of different computation platforms have a plurality of different hardware configurations of system components, respectively;

wherein the workflow execution engine is configured to select, from the plurality of different computation platforms, a distributed computation platform to execute at least part of the experiment workflow in accordance with the DG;

wherein the workflow execution engine is configured to generate, based on a hardware configuration of system components associated with the distributed computation platform, an execution schedule that specifies which system component executes each part of the classification experiment; and

wherein the workflow execution engine is configured to schedule the distributed computation platform to execute the classification experiment by the system components under the execution schedule according to the input data space and the workflow configuration, wherein the system components are scheduled to execute their respective parts of the classification experiment as specified by the execution schedule.