Identifying optimum times at which to retrain a logistic regression model

1. A method implemented by an information handling system that includes a memory and a processor, the method comprising:

generating, by the processor, a first hyperplane in response to training a machine-learning model utilizing a first set of labeled feature vectors corresponding to a first group of source documents, wherein the first set of labeled feature vectors comprises at least one yes labeled feature vector and at least one no labeled feature vector;

computing, by the processor, a first distribution of a first set of feature vectors relative to the first hyperplane, wherein the first set of feature vectors correspond to the first group of source documents;

ingesting, by the processor, a second group of source documents in response to the generation of the first hyperplane;

computing, by the processor, a second distribution of a second set of feature vectors relative to the first hyperplane, wherein the second set of feature vectors correspond to at least the second group of source documents; and

retraining, by the processor, the machine-learning model in response to determining that a distribution difference between the second distribution and the first distribution reaches a distribution difference threshold.

2. The method of claim 1 further comprising:

labeling a portion of the second set of feature vectors, resulting in a second set of labeled feature vectors; and

utilizing a combination of the first set of labeled feature vectors and the second set of labeled feature vectors to retrain the machine-learning model.

3. The method of claim 2 further comprising:

generating a second hyperplane in response to the retraining of the machine-learning model, wherein the second hyperplane is different from the first hyperplane.

4. The method of claim 1 wherein the determination that the distribution difference reaches the distribution difference further comprises:

computing a baseline ratio based upon a number of yes labeled baseline feature vectors and a number of no labeled baseline feature vectors within a predetermined width from the first hyperplane, wherein the yes labeled baseline feature vectors comprise the at least one yes labeled feature vector, and wherein the no labeled baseline feature vectors comprise the at least one no labeled feature vector;

computing a subsequent ratio based upon a number of yes labeled subsequent feature vectors and a number of no labeled subsequent feature vectors within the predetermined width from the first hyperplane; and

determining that the difference between the subsequent ratio and the baseline ratio reaches the distribution difference threshold.

5. The method of claim 1 wherein the machine-learning model is a logistic regression model utilized by a question answer system to generate candidate answers to a question.

6. The method of claim 1 wherein the second set of feature vectors includes the first set of feature vectors.

7. An information handling system comprising:

one or more processors;

a memory coupled to at least one of the processors; and

a set of computer program instructions stored in the memory and executed by at least one of the processors in order to perform actions of:

generating, by one of the one the processors, a first hyperplane in response to training a machine-learning model utilizing a first set of labeled feature vectors corresponding to a first group of source documents, wherein the first set of labeled feature vectors comprises at least one yes labeled feature vector and at least one no labeled feature vector;

computing, by one of the one the processors, a first distribution of a first set of feature vectors relative to the first hyperplane, wherein the first set of feature vectors correspond to the first group of source documents;

ingesting, by one of the one the processors, a second group of source documents in response to the generation of the first hyperplane;

computing, by one of the one the processors, a second distribution of a second set of feature vectors relative to the first hyperplane, wherein the second set of feature vectors correspond to at least the second group of source documents; and

retraining, by one of the one the processors, the machine-learning model in response to determining that a distribution difference between the second distribution and the first distribution reaches a distribution difference threshold.

8. The information handling system of claim 7 wherein the one or more processors perform additional actions comprising:

labeling a portion of the second set of feature vectors, resulting in a second set of labeled feature vectors; and

utilizing a combination of the first set of labeled feature vectors and the second set of labeled feature vectors to retrain the machine-learning model.

9. The information handling system of claim 8 wherein the one or more processors perform additional actions comprising:

generating a second hyperplane in response to the retraining of the machine-learning model, wherein the second hyperplane is different from the first hyperplane.

10. The information handling system of claim 7 wherein the one or more processors perform additional actions comprising:

computing a baseline ratio based upon a number of yes labeled baseline feature vectors and a number of no labeled baseline feature vectors within a predetermined width from the first hyperplane, wherein the yes labeled baseline feature vectors comprise the at least one yes labeled feature vector, and wherein the no labeled baseline feature vectors comprise the at least one no labeled feature vector;

computing a subsequent ratio based upon a number of yes labeled subsequent feature vectors and a number of no labeled subsequent feature vectors within the predetermined width from the first hyperplane; and

determining that the difference between the subsequent ratio and the baseline ratio reaches the distribution difference threshold.

11. The information handling system of claim 7 wherein the machine-learning model is a logistic regression model utilized by a question answer system to generate candidate answers to a question.

12. The information handling system of claim 7 wherein the second set of feature vectors includes the first set of feature vectors.

13. A computer program product stored in a computer readable storage medium, comprising computer program code that, when executed by an information handling system, causes the information handling system to perform actions comprising:

generating a first hyperplane in response to training a machine-learning model utilizing a first set of labeled feature vectors corresponding to a first group of source documents, wherein the first set of labeled feature vectors comprises at least one yes labeled feature vector and at least one no labeled feature vector;

computing a first distribution of a first set of feature vectors relative to the first hyperplane, wherein the first set of feature vectors correspond to the first group of source documents;

ingesting a second group of source documents in response to the generation of the first hyperplane;

computing a second distribution of a second set of feature vectors relative to the first hyperplane, wherein the second set of feature vectors correspond to at least the second group of source documents; and

retraining the machine-learning model in response to determining that a distribution difference between the second distribution and the first distribution reaches a distribution difference threshold.

14. The computer program product of claim 13 wherein the information handling system performs additional actions comprising:

labeling a portion of the second set of feature vectors, resulting in a second set of labeled feature vectors; and

utilizing a combination of the first set of labeled feature vectors and the second set of labeled feature vectors to retrain the machine-learning model.

15. The computer program product of claim 14 wherein the information handling system performs additional actions comprising:

generating a second hyperplane in response to the retraining of the machine-learning model, wherein the second hyperplane is different from the first hyperplane.

16. The computer program product of claim 13 wherein the information handling system performs additional actions comprising:

computing a baseline ratio based upon a number of yes labeled baseline feature vectors and a number of no labeled baseline feature vectors within a predetermined width from the first hyperplane, wherein the yes labeled baseline feature vectors comprise the at least one yes labeled feature vector, and wherein the no labeled baseline feature vectors comprise the at least one no labeled feature vector;

computing a subsequent ratio based upon a number of yes labeled subsequent feature vectors and a number of no labeled subsequent feature vectors within the predetermined width from the first hyperplane; and

determining that the difference between the subsequent ratio and the baseline ratio reaches the distribution difference threshold.

17. The computer program product of claim 13 wherein the machine-learning model is a logistic regression model utilized by a question answer system to generate candidate answers to a question.