Detecting instabilities in time series forecasting
First Claim
Patent Images
1. A data monitoring system embodied on a computer readable storage medium that comprises the following computer-executable components:
- a receiver component that receives a plurality of predictive samples from a predictive model created by way of forward sampling of one or more previously made predicted samples from the predictive model, the forward sampling creating the one or more previously made predicted samples by;
generating, by the predictive model, a first probability distribution to make a first prediction for a variable at a first instance in time in a future;
randomly drawing a sample from the probability distribution that is treated as an observed value for the variable at the first instance in time;
using the observed value to create a second probability distribution for the variable at a second instance in time in the future; and
continuing the generating, the randomly drawing, and the using the observed value until a subsequent probability distribution is obtained for the variable at a predetermined instance in time in the future; and
an analysis component that analyzes a plurality of the received predictive samples and automatically determines whether the predictive model is reliable at a time range associated with the plurality of received predictive samples, the determination is based at least in part upon a rate of change of divergence of a forward sampling operator (FS) associated with the predictive model based upon the plurality of received predictive samples, the determination further analyzing the rate of change of divergence with respect to a threshold, wherein the analysis component;
determines that the predictive model is outputting reliable predictions at a certain instance of time in the future and allowing the predictive model to continue outputting predictions when the rate of change of divergence is below the threshold;
determines that the predictive model is not outputting reliable predictions at a certain instance of time in the future and halts the predictive model from outputting predictions subsequent to that certain instance of time when the rate of change of divergence is above the threshold; and
determines that the predictive model is outputting reliable predictions at a first instance of time in the future and allowing the predictive model to continue outputting predictions when the rate of change of divergence is above the threshold at the first instance of time and when the rate of change of divergence is below the threshold at a certain consecutive number of instances of time subsequent to the first instance of time.
2 Assignments
0 Petitions
Accused Products
Abstract
A predictive model analysis system comprises a receiver component that receives predictive samples created by way of forward sampling. An analysis component analyzes a plurality of the received predictive samples and automatically determines whether a predictive model is reliable at a time range associated with the plurality of predictive sample, wherein the determination is made based at least in part upon an estimated norm associated with a forward sampling operator.
47 Citations
20 Claims
-
1. A data monitoring system embodied on a computer readable storage medium that comprises the following computer-executable components:
-
a receiver component that receives a plurality of predictive samples from a predictive model created by way of forward sampling of one or more previously made predicted samples from the predictive model, the forward sampling creating the one or more previously made predicted samples by; generating, by the predictive model, a first probability distribution to make a first prediction for a variable at a first instance in time in a future; randomly drawing a sample from the probability distribution that is treated as an observed value for the variable at the first instance in time; using the observed value to create a second probability distribution for the variable at a second instance in time in the future; and continuing the generating, the randomly drawing, and the using the observed value until a subsequent probability distribution is obtained for the variable at a predetermined instance in time in the future; and an analysis component that analyzes a plurality of the received predictive samples and automatically determines whether the predictive model is reliable at a time range associated with the plurality of received predictive samples, the determination is based at least in part upon a rate of change of divergence of a forward sampling operator (FS) associated with the predictive model based upon the plurality of received predictive samples, the determination further analyzing the rate of change of divergence with respect to a threshold, wherein the analysis component; determines that the predictive model is outputting reliable predictions at a certain instance of time in the future and allowing the predictive model to continue outputting predictions when the rate of change of divergence is below the threshold; determines that the predictive model is not outputting reliable predictions at a certain instance of time in the future and halts the predictive model from outputting predictions subsequent to that certain instance of time when the rate of change of divergence is above the threshold; and determines that the predictive model is outputting reliable predictions at a first instance of time in the future and allowing the predictive model to continue outputting predictions when the rate of change of divergence is above the threshold at the first instance of time and when the rate of change of divergence is below the threshold at a certain consecutive number of instances of time subsequent to the first instance of time. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15)
-
-
16. A computer-implemented method for determining operability of a predictive model with respect to predicted values of variables at future instances in time, comprising the following computer-executable acts:
-
receiving predictive values for a variable from a predictive model that are determined based upon forward sampling, wherein forward sampling comprises employing at least one previously predicted value from the predictive model as an observed value in predicting one or more additional values for the variable, the forward sampling creating the at least one previously predicted value by; generating, by the predictive model, a first probability distribution to make a first prediction for the variable at a first instance in time in a future; randomly drawing a sample from the probability distribution that is treated as the observed value for the variable at the first instance in time; using the observed value to create a second probability distribution for the variable at a second instance in time in the future; and continuing the generating, the randomly drawing, and the using the observed value until a subsequent probability distribution is obtained for the variable at a predetermined instance in time in the future; and automatically determining a distance in time in the future that the predictive model will reliably predict values for the variable based at least in part upon an estimated rate of change of divergence of the received predictive values the automatically determining further comprising; determining that the predictive model is outputting reliable predictions at a certain instance of time in the future and allowing the predictive model to continue outputting predictions when the rate of change of divergence is below the threshold; determining that the predictive model is not outputting reliable predictions at a certain instance of time in the future and halts the predictive model from outputting predictions subsequent to that certain instance of time when the rate of change of divergence is above the threshold; and determining that the predictive model is outputting reliable predictions at a first instance of time in the future and allowing the predictive model to continue outputting predictions when the rate of change of divergence is above the threshold at the first instance of time and when the rate of chance of divergence is below the threshold at a certain consecutive number of instances of time subsequent to the first instance of time. - View Dependent Claims (17, 18, 19)
-
-
20. A system embodied on a computer readable storage medium that facilitates monitoring of a predictive model, comprising:
-
computer-implemented means for receiving predicted values for at least one variable, the received predicted values obtained by way of forward sampling, wherein forward sampling comprises employing at least one previously predicted value from the predictive model as an observed value in determining one or more future predicted values for the at least one variable, the forward sampling creating the at least one previously predicted value by; generating, by the predictive model, a first probability distribution to make a first prediction for the at least one variable at a first instance in time in a future; randomly drawing a sample from the probability distribution that is treated as the observed value for the at least one variable at the first instance in time; using the observed value to create a second probability distribution for the at least one variable at a second instance in time in the future; and continuing the generating, the randomly drawing, and the using the observed value until a subsequent probability distribution is obtained for the at least one variable at a predetermined instance in time in the future; computer-implemented means for estimating a rate of change of divergence of a forward sampling operator (FS) associated with the predictive model based at least in part upon the received predicted values, wherein the rate of change of divergence is determined by an expectation of |FS′
|, where E(|FS′
|)=Σ
|FS′
|s/N, where E(|FS′
|) is an expectation of |FS′
| and N is a number of the received predicted samples; andcomputer-implemented means for determining a received predicted value from the predictive model that is not accurate based at least in part upon the estimated rate of change of divergence exceeding a predefined threshold, the computer-implemented means for determining further comprising; determining that the predictive model is outputting reliable predictions at a certain instance of time in the future and allowing the predictive model to continue outputting predictions when the estimated rate of change of divergence is below the predefined threshold; determining that the predictive model is not outputting reliable predictions at a certain instance of time in the future and halts the predictive model from outputting predictions subsequent to that certain instance of time when the estimated rate of change of divergence is above the predefined threshold; and determining that the predictive model is outputting reliable predictions at a first instance of time in the future and allowing the predictive model to continue outputting predictions when the estimated rate of change of divergence is above the predefined threshold at the first instance of time and when the estimated rate of change of divergence is below the predefined threshold at a certain consecutive number of instances of time subsequent to the first instance of time.
-
Specification