Software path prediction via machine learning

Abstract

Disclosed are a computer system, a software path prediction computer, non-transitory computer-readable medium, and method for determining a predicted software path that utilize segmentation machine learning in combination with ensemble machine learning to keep a most accurate model running on a server program that receives requests from and sends predicted software path(s) to a software client.

What is claimed is: 
     
         1 . A method comprising:
 performing a segmentation machine learning model on a time segmented data batch (TSDB) to obtain a segmentation output;   performing each of a plurality of machine learning models in an ensemble machine learning model on an input data set comprising the TSDB and the segmentation output;   selecting a first machine learning model from the plurality of machine learning models in the ensemble machine learning model as a most accurate model based on the step of performing each of the plurality of machine learning models in the ensemble machine learning model on an input data set;   loading the first machine learning model onto a server program as the most accurate model; and   performing, by the server program, the most accurate model on a feature set to obtain an output comprising a predicted software path for a software client.   
     
     
         2 . The method of  claim 1 , further comprising:
 receiving a stream of time series real-time software event data (TSRTSED);   compressing the TSRTSED to form a compressed TSRTSED;   storing, in a first datastore, the compressed TSRTSED;   retrieving, from the first datastore, the TSDB comprising the compressed TSRTSED that is stored over a period of time; and   sending the TSDB to the segmentation machine learning model and to the ensemble machine learning model.   
     
     
         3 . The method of  claim 2 , wherein the step of compressing compresses the received TSRTSED by greater than 50 to form the compressed TSRTSED. 
     
     
         4 . The method of  claim 3 , wherein a ratio of a size of the compressed TSRTSED to a size of the TSRTSED is less than 0.5:1. 
     
     
         5 . The method of  claim 2 , further comprising:
 storing, in a second datastore without compression, the TSRTSED;   retrieving, by the server program, the TSRTSED from the second datastore;   building, by the server program, the feature set from the TSRTSED retrieved from the second datastore, wherein the feature set is unique to a user of the software client; and   inputting, by the server program, the feature set to the most accurate model loaded thereon.   
     
     
         6 . The method of  claim 2 , further comprising:
 sending, by the server program, a feedback containing the predicted software path to a data ingestion computer that is configured to include the feedback in a time series real-time software event data (TSRTSED).   
     
     
         7 . The method of  claim 1 , further comprising:
 receiving a request for the predicted software path from the software client, wherein the request comprises an identifier unique to a user of the software client; and   sending the predicted software path to the software client.   
     
     
         8 . The method of  claim 1 , wherein the segmentation machine learning model is configured to segment the TSDB into a plurality of user groups to obtain the segmentation output. 
     
     
         9 . The method of  claim 1 , further comprising:
 storing software path options on a third datastore;   wherein the input data set further comprises the software path options.   
     
     
         10 . The method of  claim 1 , further comprising:
 determining a second machine learning model of the plurality of machine learning models in the ensemble machine learning model is more accurate than the first machine learning model;   identifying the second machine learning model as the most accurate model based on the step of determining;   updating the server program with the second machine learning model as the most accurate model loaded on the server program; and   performing, by the server program, the second machine learning model as the most accurate model on another feature set to obtain another output comprising another predicted software path for the software client or another software client.   
     
     
         11 . The method of  claim 10 , wherein the step of determining comprises:
 performing the segmentation machine learning model on a second time segmented data batch (TSDB) to obtain a second segmentation output;   performing each of the plurality of machine learning models in the ensemble machine learning model on a second input data set comprising the second TSDB and the second segmentation output; and   determining an accuracy of the first machine learning model loaded on the server program is less than an accuracy of the second machine learning model of the plurality of machine learning models in the ensemble machine learning model.   
     
     
         12 . The method of  claim 1 , wherein the server program is periodically updated to perform a most recently accurate model on additional feature sets. 
     
     
         13 . A computer system comprising:
 a software path prediction computer having a first datastore, a second datastore, and a server program running thereon, wherein the software path prediction computer is configured to: 
 perform a segmentation machine learning model on a time segmented data batch (TSDB) to obtain a segmentation output; 
 perform each of a plurality of machine learning models in an ensemble machine learning model on an input data set comprising the TSDB and the segmentation output; 
 select a first machine learning model from the plurality of machine learning models in the ensemble machine learning model as a most accurate model based on the step of performing each of the plurality of machine learning models in the ensemble machine learning model on an input data set; and 
 load the first machine learning model onto a server program as the most accurate model; 
 wherein the server program is configured to perform the most accurate model on a feature set to obtain an output comprising a predicted software path for a software client. 
   
     
     
         14 . The computer system of  claim 13 , wherein the software path prediction computer is further configured to:
 receive a stream of time series real-time software event data (TSRTSED);   compress the TSRTSED to form a compressed TSRTSED;   store, in the first datastore, the compressed TSRTSED;   retrieve, from the first datastore, the TSDB comprising the compressed TSRTSED that is stored over a period of time; and   send the TSDB to the segmentation machine learning model and to the ensemble machine learning model.   
     
     
         15 . The computer system of  claim 14 , wherein the software path prediction computer is further configured to:
 store, in the second datastore without compression, the TSRTSED;   wherein the server program is further configured to:   retrieve the TSRTSED from the second datastore;   build the feature set from the TSRTSED retrieved from the second datastore, wherein the feature set is unique to a user of the software client; and   input the feature set to the most accurate model loaded thereon.   
     
     
         16 . The computer system of  claim 13 , wherein the software path prediction computer is further configured to:
 determine a second machine learning model of the plurality of machine learning models in the ensemble machine learning model is more accurate than the first machine learning model;   identify the second machine learning model as the most accurate model based on the step of determining; and   update the server program with the second machine learning model as the most accurate model loaded on the server program;   wherein the server program is further configured to perform the second machine learning model as the most accurate model on another feature set to obtain another output comprising another predicted software path for the software client or another software client.   
     
     
         17 . The computer system of  claim 13 , wherein the software path prediction computer is networked with a device running the software client, wherein the server program of the software path prediction computer is configured to receive a request for the predicted software path from the device and to send the predicted software path to the device. 
     
     
         18 . The computer system of  claim 13 , further comprising:
 a data ingestion computer networked with the software path prediction computer, wherein the data ingestion computer is configured to receive a stream of software event data, to convert the software event data to time series real-time software event data (TSRTSED), and to send the TSRTSED to the software path prediction computer.   
     
     
         19 . A non-transitory computer-readable medium storing computer readable instructions that, when executed by one or more computing devices, causes the one or more computing devices to perform operations comprising:
 performing a segmentation machine learning model on a time segmented data batch (TSDB) to obtain a segmentation output;   performing each of a plurality of machine learning models in an ensemble machine learning model on an input data set comprising the TSDB and the segmentation output;   selecting a first machine learning model from the plurality of machine learning models in the ensemble machine learning model as a most accurate model based on the step of performing each of the plurality of machine learning models in the ensemble machine learning model on an input data set;   loading the first machine learning model onto a server program as the most accurate model; and   performing, by the server program, the most accurate model on a feature set to obtain an output comprising a predicted software path for a software client.   
     
     
         20 . The non-transitory computer-readable medium of  claim 19 , wherein the operations further comprise:
 receiving a stream of time series real-time software event data (TSRTSED);   compressing the TSRTSED to form a compressed TSRTSED;   storing, in the first datastore, the compressed TSRTSED;   retrieving, from the first datastore, the TSDB comprising the compressed TSRTSED that is stored over a period of time;   sending the TSDB to the segmentation machine learning model and to the ensemble machine learning model;   storing, in the second datastore without compression, the TSRTSED;   retrieving the TSRTSED from the second datastore;   building the feature set from the TSRTSED retrieved from the second datastore, wherein the feature set is unique to a user of the software client; and   inputting the feature set to the most accurate model.