Dataset suitability check for machine learning

1. A device comprising:

a processing device; and

a memory to store computer executable instructions that, if executed, cause the processing device to:

perform a plurality of k-nearest neighbors search queries with various values of k on a dataset, wherein the dataset comprises a plurality of data points, and wherein each data point of the plurality of data points comprises a representative vector for the data point and an associated classification for the data point, wherein a first set of the plurality of data points comprises a first classification and a second set of the plurality of data points comprises a second classification;

determine a score representative of a degree of clustering of the plurality of data points from results of the plurality of k-nearest neighbors search queries, wherein the score is determined based on a first degree of clustering of data points classified with the first classification of the first set of the plurality of data points and a second degree of clustering of data points classified with the second classification of the second set of the plurality of data points; and

determine a suitability of the dataset for use in generating a k-nearest neighbors search machine learning model based on a comparison of the score to a threshold.

2. The device of claim 1 , wherein the instructions, if executed, further cause the processing device to:

determine a plurality of scores, wherein each of the plurality of scores is determined using a different value of k for a k-nearest neighbors search of the plurality of k-nearest neighbors search queries;

determine a value of k that resulted in a highest score from the plurality of scores; and

determine that the value of k that resulted in the highest score is an optimal value of k for the k-nearest neighbors search.

3. The device of claim 1 , wherein the instructions, if executed, further cause the processing device to:

compute a plurality of precision-recall curves, wherein a separate precision-recall curve is computed for two or more data points of the plurality of data points based on results of a plurality of k-nearest neighbors search queries associated with the two or more data points;

for each precision-recall curve of the plurality of precision-recall curves, compute an area under the precision-recall curve; and

compute an average area under the plurality of precision-recall curves, wherein the score representative of the degree of clustering of the plurality of data points is based on the average area under the plurality of precision-recall curves.

4. The device of claim 1 , wherein the instructions, if executed, further cause the processing device to:

determine one or more first precision-recall curves having a highest area;

determine one or more second precision-recall curves having a lowest area; and

discard the one or more first precision-recall curves and the one or more second precision-recall curves.

5. The device of claim 1 , wherein the instructions, if executed, further cause the processing device to:

perform a number of k-nearest neighbors search queries for a data point of the plurality of data points, where the number is equal to s−1, where s is a size of the dataset, and where each of the number of k-nearest neighbors search queries uses a different value of k that is between 1 and s−1, inclusive.

6. The device of claim 1 , wherein the instructions, if executed, further cause the processing device to:

receive a plurality of additional data points for the dataset, wherein addition of the plurality of additional data points to the dataset results in an updated dataset;

determine a new score for the updated dataset;

compare the score to the new score; and

determine whether the new score is greater than the score.

7. A method comprising:

receiving a dataset comprising a plurality of data points, wherein each data point of the plurality of data points comprises a representative vector for the data point and an associated classification for the data point, wherein a first set of the plurality of data points comprises a first classification and a second set of the plurality of data points comprises a second classification;

determining, for the dataset, a score representative of a degree of clustering of the plurality of data points from results of a plurality of k-nearest neighbors search queries, wherein the score is determined based on a first degree of clustering of data points classified with the first classification of the first set of the plurality of data points and a second degree of clustering of data points classified with the second classification of the second set of the plurality of data points; and

determining a suitability of the dataset for use in machine learning based on a comparison of the score to a threshold.

8. The method of claim 7 , wherein the suitability of the dataset is determined for a k-nearest neighbors search.

9. The method of claim 8 , further comprising:

determining a plurality of scores, wherein each of the plurality of scores is determined using a different value of k for the k-nearest neighbors search;

determining a value of k that resulted in a highest score from the plurality of scores; and

determining that the value of k that resulted in the highest score is an optimal value of k for the k-nearest neighbors search.

10. The method of claim 7 , further comprising:

performing a plurality of k-nearest neighbors search queries with various values of k on the dataset, wherein the score is determined based on combined results of the plurality of k-nearest neighbors search queries.

11. The method of claim 10 , further comprising:

computing a plurality of precision-recall curves, wherein a separate precision-recall curve is computed for two or more data points of the plurality of data points based on results of a plurality of k-nearest neighbors search queries associated with the two or more data points;

for each precision-recall curve of the plurality of precision-recall curves, computing an area under the precision-recall curve; and

computing an average area under the plurality of precision-recall curves, wherein the score representative of the degree of clustering of the plurality of data points is based on the average area under the plurality of precision-recall curves.

12. The method of claim 11 , further comprising:

determining one or more first precision-recall curves having a highest area;

determining one or more second precision-recall curves having a lowest area; and

discarding the one or more first precision-recall curves and the one or more second precision-recall curves.

13. The method of claim 10 , wherein performing the plurality of k-nearest neighbors search queries comprises:

performing a number of k-nearest neighbors search queries for a data point of the plurality of data points using various values of k between a and s−1, inclusive, where s is a size of the dataset.

14. The method of claim 13 , wherein performing the plurality of k-nearest neighbors search queries further comprises performing the number of k-nearest neighbors search queries for each additional data point of the plurality of data points.

15. The method of claim 7 , further comprising:

comparing the score to a threshold;

determining that the score meets or exceeds the threshold; and

determining that the dataset is suitable for use in machine learning.

16. The method of claim 7 , further comprising:

receiving a plurality of additional data points for the dataset to obtain an updated dataset;

determining a new score for the updated dataset;

comparing the score to the new score; and

determining whether the new score is greater than the score.

17. A non-transitory computer-readable storage device storing computer-executable instructions that, if executed by a processing device, cause the processing device to:

perform a plurality of search queries on a dataset, wherein the dataset comprises a plurality of data points, wherein each data point of the plurality of data points comprises a representative vector for the data point and an associated classification for the data point, and wherein a first set of the plurality of data points comprises a first classification and a second set of the plurality of data points comprises a second classification;

determine a score representative of a degree of clustering of the plurality of data points from results of the plurality of search queries, wherein the score is determined based on a first degree of clustering of data points classified with the first classification of the first set of the plurality of data points and a second degree of clustering of data points classified with the second classification of the second set of the plurality of data points; and

determine a suitability of the dataset for use in machine learning based on a comparison of the score to a threshold.

18. The non-transitory computer-readable storage device of claim 17 , wherein the machine learning comprises a k-nearest neighbors search, and wherein the plurality of search queries comprise a plurality of k-nearest neighbors search queries with various values of k.

19. The non-transitory computer-readable storage device of claim 18 , wherein the instructions, if executed, further cause the processing device to:

compute a plurality of precision-recall curves, wherein a separate precision-recall curve is computed for two or more data points of the plurality of data points based on results of k-nearest neighbors search queries from the plurality of k-nearest neighbors search queries that are associated with the two or more data points;

for each precision-recall curve of the plurality of precision-recall curves, compute an area under the precision-recall curve; and

compute an average area under the plurality of precision-recall curves, wherein the score representative of the degree of clustering of the plurality of data points is based on the average area under the plurality of precision-recall curves.

20. The non-transitory computer-readable storage device of claim 19 , wherein performing the plurality of k-nearest neighbors search queries comprises:

performing a number of k-nearest neighbors search queries for a data point of the plurality of data points using various values of k between 1 and s−1, inclusive, where s is a size of the dataset; and

performing the number of k-nearest neighbors search queries for each additional data point of the plurality of data points.