Systems for time-series predictive data analytics, and related methods and apparatus

Patent metadata

Field	Value
Publication number	US-12423595-B2
Application number	US-201916682813-A
Country	US
Kind code	B2
Filing date	Nov 13, 2019
Priority date	May 23, 2014
Publication date	Sep 23, 2025
Grant date	Sep 23, 2025

Abstract

A predictive modeling method may include determining a time interval of time-series data; identifying one or more variables of the data as targets; determining a forecast range and a skip range associated with a prediction problem represented by the data; generating training data and testing data from the time-series data; fitting a predictive model to the training data; and testing the fitted model on the testing data. The forecast range may indicate a duration of a period for which values of the targets are to be predicted. The skip range may indicate a temporal lag between the time period corresponding to the data used to make predictions and the time period corresponding to the predictions. The skip range may separate input data subsets representing model inputs from subsets representing model outputs, and separate test data subsets representing model inputs from subsets representing validation data.

First claim

What is claimed is:

1. A computer-implemented modeling method comprising:
obtaining a plurality of data sets, wherein each data set includes a plurality of observations, wherein each observation includes an indication of a time of the observation and values of one or more variables, wherein at least one of the variables is a target;
setting values of a plurality of temporal parameters of a modeling procedure, the plurality of temporal parameters including a forecast range parameter and a skip range parameter, wherein the forecast range parameter indicates a duration of a period of time for prediction of one or more values of the target, and wherein the skip range parameter indicates a duration of a time period between a time of an earliest-in-time prediction in the forecast range and a time of a latest-in-time observation upon which the earliest-in-time prediction in the forecast range is based;
segmenting at least one of the data sets into training-input data and training-output data based, at least in part, on the values of the forecast range parameter and the skip range parameter, wherein the training-input data include a first subset of the observations of the at least one data set, the training-output data include a second subset of the observations of the at least one data set, a range of the times of the observations in the second subset matches the value of the forecast range parameter, and a range from a latest-in-time observation in the first subset to an earliest-in-time observation in the second subset matches the value of the skip range parameter; and
adapting a model to solve a prediction problem represented by the plurality of data sets and the values of the forecast range and skip range parameters, including training the model using the training-input data and the training-output data.

2. The method of claim 1 , wherein at least a group of the observations of the plurality of data sets include respective values of a first variable, and wherein the method further comprises, prior to training the model using the training-input data and the training-output data:
determining that the values of the first variable comprise time values;
for each observation in the group, generating a respective value of a second variable, wherein the respective value of the second variable comprises an offset between the respective time value of the first variable and a reference time value; and
adding the respective values of the second variable to the respective observations in the group.

3. The method of claim 2 , further comprising: removing the values of the first variable from the observations in the group.

4. The method of claim 1 , further comprising determining the value of the forecast range parameter based, at least in part, on (1) a time interval of the plurality of data sets, (2) a number of observations included in the plurality of data sets, (3) a time period corresponding to the plurality of data sets, and/or (4) a natural time period selected from the group consisting of microseconds, milliseconds, seconds, minutes, hours, days, weeks, months, quarters, seasons, years, decades, centuries, and millennia.

5. The method of claim 1 , further comprising determining the value of the skip range parameter based, at least in part, on latency in collection of the plurality of data sets, latency in communication of the plurality of data sets, latency in analyzing the plurality of data sets, latency in communication of analyses of the plurality of data sets and/or latency of implementing actions based on the analyses of the plurality of data sets.

6. The method of claim 1 , further comprising:
selecting validation data from the plurality of data sets based on the value of the forecast range parameter and the value of the skip range parameter, wherein the validation data include a subset of observations of at least one of the data sets; and
performing cross-validation of the model using the validation data.

7. The method of claim 6 , wherein the model is trained and/or cross-validated using sliding windows, wherein the sliding windows correspond to sequential subsets of observations from at least one of the data sets.

8. The method of claim 1 , further comprising:
selecting validation data from the plurality of data sets based on the value of the forecast range parameter and the value of the skip range parameter, wherein the validation data include a subset of observations of at least one of the data sets; and
performing nested cross-validation of the model using the validation data.

9. The method of claim 1 , wherein the indication of the time of each observation comprises a time at which the respective observation was made or another time associated with the respective observation.

10. The method of claim 1 , wherein the one or more variables include a first variable and a second variable, and wherein the method further comprises:
determining that changes in the values of the first and second variables are correlated with a duration of a time period between the changes in the value of the first variable and the correlated changes in the value of the second variable; and
displaying, via a graphical user interface, graphical content indicating the duration of the time period between the changes in the value of the first variable and the correlated changes in the value of the second variable.

11. The method of claim 1 , further comprising:
determining suitabilities of a plurality of modeling procedures for the prediction problem based, at least in part, on characteristics of the prediction problem and/or on attributes of the respective modeling procedures;
selecting one or more modeling procedures from the plurality of modeling procedures based on the determined suitabilities of the selected modeling procedures for the prediction problem; and
performing the one or more modeling procedures.

12. The method of claim 11 , wherein performing the one or more modeling procedures comprises:
transmitting instructions to a plurality of processing nodes, the instructions comprising a resource allocation schedule allocating resources of the processing nodes for execution of the one or more modeling procedures, the resource allocation schedule being based, at least in part, on the suitabilities of the one or more modeling procedures for the prediction problem;
receiving results of the execution of the one or more modeling procedures by the plurality of processing nodes in accordance with the resource allocation schedule, wherein the results include one or more models generated by the one or more modeling procedures, and/or scores of the generated one or more models for data associated with the prediction problem; and
selecting, from the generated one or more models, a particular model based, at least in part, on the score of the particular model.

13. The method of claim 1 , further comprising deploying the trained model, wherein the trained model is deployed to one or more servers, wherein other trained models are also deployed to the one or more servers, and wherein requests to the trained model and the other trained models are allocated among the servers based, at least in part, on (1) an estimate of an amount of time used by each of the trained models to generate a response, and/or (2) an estimate of a frequency with which requests for each of the trained models are received.

14. The method of claim 13 , wherein each request is assigned to a respective thread, wherein each request has an associated latency-sensitivity value, and wherein a number of threads executing on a particular server is determined based, at least in part, on the lat

Assignees

• Datarobot Inc

Inventors

• Achin Jeremy
• Degodoy Thomas
• Conort Xavier
• Steadman Mark L
• Prettenhofer Peter
• Owen Timothy

Classifications

• Y02P90/80

  Management or planning · CPC title

• G06N20/20

  Ensemble learning · CPC title

• G06N20/10

  using kernel methods, e.g. support vector machines [SVM] · CPC title

• G06N5/046

  Forward inferencing; Production systems · CPC title

• G06N20/00 Primary

  Machine learning · CPC title