Pipeline system for time-series data forecasting

Patent metadata

Field	Value
Publication number	US-10560313-B2
Application number	US-201916452791-A
Country	US
Kind code	B2
Filing date	Jun 26, 2019
Priority date	Jun 26, 2018
Publication date	Feb 11, 2020
Grant date	Feb 11, 2020

Abstract

A pipeline system for time-series data forecasting using a distributed computing environment is disclosed herein. In one example, a pipeline for forecasting time series is generated. The pipeline represents a sequence of operations for processing the time series to produce forecasts. The sequence of operations include model strategy operations for applying various model strategies to the time series to determine error distributions corresponding to the model strategies. The sequence of operations further include a model-strategy comparison operation for determining which of the model strategies is a champion model strategy for the plurality of time series based on the error distributions of the model strategies. The pipeline is executed to determine the champion model strategy for the time series.

First claim

The invention claimed is:

1. A system comprising:
a processing device; and
a memory device comprising instructions that are executable by the processing device for causing the processing device to:
access a pipeline for forecasting a plurality of time series, the pipeline representing a sequence of operations for processing the plurality of time series to produce forecasts, wherein the sequence of operations includes:
(i) model strategy operations for applying model strategies to the plurality of time series to determine error distributions corresponding to the model strategies, wherein each of the model strategies is configured to:
receive the plurality of time series;
determine champion forecasts for the plurality of time series by, for each time series among the plurality of time series:
 generating respective forecasts for the time series by applying forecasting models to the time series; and
 determining a respective champion forecast for the time series by comparing error metrics associated with the respective forecasts;
determine error values associated with the champion forecasts, each of the error values indicating the accuracy of a respective champion forecast among the champion forecasts; and
generate one of the error distributions using the error values; and

(ii) a model-strategy comparison operation configured to:
receive the error distributions corresponding to the model strategies;
determine aggregate error values associated with the model strategies based on the error distributions, each aggregate error value being associated with one of the model strategies and being determined by aggregating the error values in one of the error distributions; and
determine which of the model strategies is a champion model strategy for the plurality of time series by comparing the aggregate error values to one another; and

execute the pipeline to determine the champion model strategy for the plurality of time series.

2. The system of claim 1 , wherein the model strategies are defined in shareable files to enable importation of the model strategies into another pipeline for execution.

3. The system of claim 1 , wherein the pipeline is programmatically generated according to user preferences within a forecasting software environment.

4. The system of claim 1 , wherein the pipeline includes:
a plurality of pipeline segments for executing a plurality of operation groups in parallel;
a segmentation operation prior to the plurality of pipeline segments for dividing the plurality of time series into a plurality of data segments, each data segment among the plurality of data segments being destined for a respective pipeline segment among the plurality of pipeline segments; and
a merge segment operation subsequent to the pipeline segments for combining results from the plurality of pipeline segments.

5. The system of claim 1 , wherein the memory device further comprises instructions that are executable by the processing device for causing the processing device to determine a champion pipeline for a project by comparing a first pipeline in the project to a second pipeline in the project.

6. The system of claim 5 , wherein the memory device further includes instructions that are executable by the processing device for causing the processing device to:
determine the champion pipeline for the project by comparing (i) a first champion forecast generated by the first pipeline based on the plurality of time series, to (ii) a second champion forecast generated by the second pipeline based on the plurality of time series; and
include a graphical object in a graphical user interface to indicate that the first pipeline is the champion pipeline.

7. The system of claim 6 , wherein the memory device further includes instructions that are executable by the processing device for causing the processing device to:
receive an override instruction to use the second pipeline rather than the champion pipeline to forecast the plurality of time series; and
subsequent to receiving the override instruction, generate one or more forecasts for the plurality of time series using the second pipeline rather than the champion pipeline.

8. The system of claim 1 , wherein:
the pipeline is defined in a file that is a shareable with other users and formatted to enable the other users to import the pipeline into a forecasting software environment and execute the pipeline;
accessing the pipeline comprises loading the file of the pipeline into the forecasting software environment; and
executing the pipeline comprises executing the pipeline in the forecasting software environment.

9. The system of claim 1 , wherein the memory device further includes instructions that are executable by the processing device for causing the processing device to:
include a graphical visualization of the pipeline in a graphical user interface, the graphical visualization including visual representations of each operation among the sequence of operations spatially organized in an order in which the sequence of operations is executed in the pipeline.

10. The system of claim 9 , wherein the graphical visualization includes a swim lane diagram with a series of distinct lanes spatially positioned adjacent to one another, the lanes including:
a first lane for depicting pre-processing operations;
a second lane for depicting model strategy operations;
a third lane for depicting comparison operations; and
a fourth lane for depicting post-processing operations,
wherein each lane has a visual representation of an operation in the pipeline connected to another visual representation of another operation in the pipeline depicted in another lane.

11. The system of claim 9 , wherein the memory device further comprises instructions that are executable by the processing device for causing the processing device to:
detect a user input involving adding an operation to the pipeline; and
based on the user input, automatically add another operation into the pipeline or remove the other operation from the pipeline in accordance with predefined rules.

12. The system of claim 11 , wherein the operation is a model strategy operation, and wherein the memory device further comprises instructions that are executable by the processing device for causing the processing device to:
automatically add the other operation to the pipeline, the other operation being a required operation according to the predefined rules; and
automatically update the graphical visualization of the pipeline to include (i) a visual representation of the other operation, and (ii) a visual representation of the model strategy operation with a visual connection to another visual representation of at least one other operation among the sequence of operations.

13. The system of claim 9 , wherein the memory device further comprises instructions that are executable by the processing device for causing the processing device to:
receive a user input involving removing an existing operation from the pipeline; and
in response to receiving the user input:
remove the existing operation from the pipeline;
automatically remove at least one other operation from the pipeline, wherein the at least one other operation is an extraneous operation based on removing the existing operation from the pipeline; and
automatically update the graphical visualization of the pipeline to remove visual representations of the existing operation and the at least one other operation.

14. A method comprising:
accessing a pipeline for forecasting a plurality of time series, the pipeline representing a sequence of operations for processing

Assignees

• Sas Inst Inc

Inventors

• Sglavo Udo Vincenzo
• Helmkamp Phillip Mark
• Brzezicki Jerzy Michal
• Haley Timothy Patrick
• Pothireddy Sujatha

Classifications

• G06N3/08

  Learning methods · CPC title

• G06F11/3452

  Performance evaluation by statistical analysis · CPC title

• G06F11/3447

  Performance evaluation by modeling · CPC title

• H04L41/16

  using machine learning or artificial intelligence · CPC title

• G06N5/04

  Inference or reasoning models · CPC title