Machine-learning models to leverage behavior-dependent processes

Abstract

Provided is a process, including: obtaining a first training dataset of subject-entity records; training a first machine-learning model on the first training dataset; forming virtual subject-entity records by appending members of a set of candidate action sequences to time-series of at least some of the subject-entity records; forming a second training dataset by labeling the virtual subject-entity records with predictions of the first machine-learning model; and training a second machine-learning model on the second training dataset.

What is claimed is: 
     
         1 . A tangible, non-transitory, machine-readable medium storing instructions that when executed by one or more processors effectuate operations comprising:
 obtaining, with a computer system, a first set of training data,
 the first set of training data comprising a time-series of events that are caused by an actor entity, and 
 at least some events of the time-series of events comprising a plurality of attributes; 
   selecting, with the computer system, a plurality of subsets of the first set of training data, a first subset among the plurality of subsets representing a first interval of time and a second subset among the plurality of subsets representing a second interval of time after the first interval of time;   training, with the computer system, a first machine-learning model on the first set of training data by optimizing parameters of the first machine-learning model with a first objective function based on an accuracy of the first machine-learning model in predicting attributes of the second subset based on the attributes of the first subset;   generating, with the computer system, a virtual set of training data, comprising:
 virtual events in a third interval after the first interval; and 
 virtual events in a fourth interval after the third interval; 
   training, with the computer system, a second machine-learning model on the virtual set of training data by optimizing parameters of the second machine-learning model with a second objective function based on an accuracy of the second machine-learning model in predicting attributes of the fourth subset based on the attributes of the third subset; and   storing, with the computer system, the trained second machine-learning model in memory.   
     
     
         2 . The medium of  claim 1 , wherein:
 the first set of training data comprise a plurality of classifications; and   the first machine-learning model is trained based on the plurality of classifications.   
     
     
         3 . The medium of  claim 1 , wherein the generation of the virtual set of training data comprises repeatedly adding new time-series of events that are caused by the actor entity. 
     
     
         4 . The medium of  claim 1 , wherein:
 the third subset comprises exogenous events.   
     
     
         5 . The medium of  claim 1 , the operations further comprising:
 filtering attributes from the second subset that are identical to the attributes of the first subset.   
     
     
         6 . The medium of  claim 1 , wherein:
 the third subset comprises events that are caused by other actor entities, wherein the events caused by other actor entities occurred before the third interval.   
     
     
         7 . The medium of  claim 1 , wherein:
 obtaining the first set of training data further comprises filtering the events that occurred before a designated date from the time-series of events.   
     
     
         8 . The medium of  claim 1 , wherein:
 at least a portion of the second interval overlaps with at least a portion of the third interval.   
     
     
         9 . The medium of  claim 1 , wherein:
 the attributes of at least some events comprise a question and a response received from the actor entity.   
     
     
         10 . The medium of  claim 1 , wherein:
 at least some events comprise offers presented to the actor entity and offers accepted by the actor entity.   
     
     
         11 . The medium of  claim 1 , the operations further comprising:
 adjusting parameters of a value function indicative of an occurrence probability of an attribute in the fourth subset.   
     
     
         12 . The medium of  claim 11 , the operations further comprising:
 scoring attributes of the fourth subset based on the occurrence probability calculated by the value function.   
     
     
         13 . The medium of  claim 12 , the operations further comprising:
 generating sequences of future events based on the score of the attributes of the fourth subset.   
     
     
         14 . The medium of  claim 1 , wherein:
 the first machine learning model is part of a value function or an environment model of a reinforcement learning model; and   the second trained machine learning model is a random decision forest model that includes a plurality of weighted trained decision trees.   
     
     
         15 . The medium of  claim 1 , wherein:
 the first plurality of subsets comprises more than 100,000 different attributes;   the first or the second machine learning model executes on a compute cluster having a plurality of computing devices that collectively perform an in-memory cluster computing;   program state upon which the first or the second machine learning model operates is stored in an in-memory, immutable, distributed dataset spread over a plurality of nodes of the compute cluster such that the distributed dataset is resilient to failure of a given one of the computing devices; and   the compute cluster concurrently processes data in the distributed dataset to apply or train the first or the second machine learning model.   
     
     
         16 . The medium of  claim 1 , wherein:
 training the first machine-learning model comprises steps for training a supervised time-series forecasting model; and   training the second machine-learning model comprises steps for training a supervised classification model.   
     
     
         17 . The medium of  claim 1 , wherein:
 the operations comprise steps for causing at least some of the attributes of the fourth plurality of the subsets to respond to the actor entity in a targeted manner based on a trained model.   
     
     
         18 . The medium of  claim 1 , wherein:
 the operations comprise steps for predicting probability of the attributes of the fourth plurality of the subsets related to the actor entity.   
     
     
         19 . The medium of  claim 1 , the operations further comprising:
 adjusting, through a plurality of iterations, parameters of the second machine-learning model to increase the accuracy of the second machine-learning model in predicting the attributes of the fourth plurality of the subsets.   
     
     
         20 . A method, comprising:
 obtaining, with a computer system, a first set of training data,
 the first set of training data comprising a time-series of events that are caused by an actor entity, and 
 at least some events of the time-series of events comprising a plurality of attributes; 
   selecting, with the computer system, a plurality of subsets of the first set of training data, a first subset among the plurality of subsets representing a first interval of time and a second subset among the plurality of subsets representing a second interval of time after the first interval of time;   training, with the computer system, a first machine-learning model on the first set of training data by optimizing parameters of the first machine-learning model with a first objective function based on an accuracy of the first machine-learning model in predicting attributes of the second subset based on the attributes of the first subset;   generating, with the computer system, a virtual set of training data, comprising:
 virtual events in a third interval after the first interval; and 
 virtual events in a fourth interval after the third interval; 
   training, with the computer system, a second machine-learning model on the virtual set of training data by optimizing parameters of the second machine-learning model with a second objective function based on an accuracy of the second machine-learning model in predicting attributes of the fourth subset based on the attributes of the third subset; and   storing, with the computer system, the trained second machine-learning model in memory.