Machine-learning techniques for evaluating suitability of candidate datasets for target applications
Techniques disclosed herein relate generally to evaluating and selecting candidate datasets for use by software applications, such as selecting candidate datasets for training machine-learning models used in software applications. Various machine-learning and other data science techniques are used to identify unique entities in a candidate dataset that are likely to be part of target entities for a software application. A merit attribute is then determined for the candidate dataset based on the number of unique entities that are likely to be part of the target entities, and weights associated with these unique entities. The merit attribute is used to identify the most efficient or most cost-effective candidate dataset for the software application.
1. A method for applying machine-learning techniques to evaluate candidate datasets for use by software applications, the method comprising performing, by one or more processing devices, operations including:
receiving (i) a reference dataset identifying first entities associated with first features that include a baseline feature of a target population and (ii) a candidate dataset identifying second entities associated with second features;
identifying, in the candidate dataset, first unique candidate entities that are absent from the reference dataset and that are associated with the baseline feature in the candidate dataset;
forming, in a multi-dimensional space and based on a subset of the first features lacking the baseline feature, a cluster of data points representing the first entities, wherein forming the cluster of data points representing the first entities comprises:
generating an aggregated matrix that identifies the first entities, the first features, the second entities, and the second features;
estimating missing values in the aggregated matrix using spectral regularization; and
mapping, based on the aggregated matrix, the first entities to the data points representing the first entities, wherein coordinates of each data point in the data points representing the first entities are determined based on corresponding values in the aggregated matrix;
mapping a subset of the second entities that are absent from the reference dataset and that are not in the first unique candidate entities to additional data points, respectively in the multi-dimensional space, wherein mapping the subset of the second entities to the additional data points in the multi-dimensional space comprises mapping each second entity in the subset of the second entities to a respective additional data point in the multi-dimensional space based on corresponding values in the aggregated matrix;
identifying, from the subset of the second entities, second unique candidate entities corresponding to a subset of the additional data points within a threshold distance of the cluster;
determining a merit attribute of the candidate dataset based on a first weight for each first unique candidate entity, a second weight for each second unique candidate entity, a number of the first unique candidate entities in the candidate dataset, and a number of the second unique candidate entities in the candidate dataset; and
selecting the candidate dataset as input data for a target software application based on the merit attribute of the candidate dataset being greater than a threshold value.
2. The method of claim 1 , wherein identifying the second unique candidate entities comprises:
determining, in the multi-dimensional space, a centroid of the cluster of the data points representing the first entities;
determining, in the multi-dimensional space, an average reference distance between each data point in the data points representing the first entities and the centroid of the cluster; and
determining the threshold distance based on the average reference distance.
3. The method of claim 2 , wherein identifying the second unique candidate entities further comprises:
determining, in the multi-dimensional space, a distance between a respective additional data point and the centroid of the cluster.
4. The method of claim 3 , wherein the distance between the respective additional data point and the centroid of the cluster includes a Pearson correlation distance, Euclidean distance, cosine distance, or Jaccard distance.
5. The method of claim 1 , wherein the operations further comprise determining, based on merit attributes of training datasets, the first weight and the second weight using linear regression.
6. The method of claim 1 , wherein determining the merit attribute of the candidate dataset comprises:
determining a weighted sum of the number of the first unique candidate entities in the candidate dataset and the number of the second unique candidate entities in the candidate dataset,
wherein each of the first unique candidate entities is associated with the first weight; and
wherein each of the second unique candidate entities is associated with the second weight.
7. A system comprising:
a processing device; and
a non-transitory computer-readable medium communicatively coupled to the processing device, wherein the processing device is configured to execute program code stored in the non-transitory computer-readable medium and thereby perform operations comprising:
receiving (i) a reference dataset identifying first entities associated with first features that include a baseline feature of a target population and (ii) a candidate dataset identifying second entities associated with second features;
identifying, in the candidate dataset, first unique candidate entities that are absent from the reference dataset and that are associated with the baseline feature in the candidate dataset;
forming, in a multi-dimensional space and based on a subset of the first features lacking the baseline feature, a cluster of data points representing the first entities, wherein forming the cluster of data points representing the first entities comprises:
generating an aggregated matrix that identifies the first entities, the first features, the second entities, and the second features;
estimating missing values in the aggregated matrix using spectral regularization; and
mapping, based on the aggregated matrix, the first entities to the data points representing the first entities, wherein coordinates of each data point in the data points representing the first entities are determined based on corresponding values in the aggregated matrix;
mapping a subset of the second entities that are absent from the reference dataset and that are not in the first unique candidate entities to additional data points, respectively in the multi-dimensional space, wherein mapping the subset of the second entities to the additional data points in the multi-dimensional space comprises mapping each second entity in the subset of the second entities to a respective additional data point in the multi-dimensional space based on corresponding values in the aggregated matrix;
identifying, from the subset of the second entities, second unique candidate entities corresponding to a subset of the additional data points within a threshold distance of the cluster;
determining a merit attribute of the candidate dataset based on a first weight for each first unique candidate entity, a second weight for each second unique candidate entity, a number of the first unique candidate entities in the candidate dataset, and a number of the second unique candidate entities in the candidate dataset; and
selecting the candidate dataset as input data for a target software application based on the merit attribute of the candidate dataset being greater than a threshold value.
8. The system of claim 7 , wherein identifying the second unique candidate entities comprises:
determining, in the multi-dimensional space, a centroid of the cluster of the data points representing the first entities;
determining, in the multi-dimensional space, an average reference distance between each data point in the data points representing the first entities and the centroid of the cluster;
determining the threshold distance based on the average reference distance; and
determining, in the multi-dimensional space, a distance between a respective additional data point and the centroid of the cluster.
9. The system of claim 8 , wherein the distance between the respective additional data point and the centroid of the cluster includes a Pearson correlation distance, Euclidean distance, cosine distance, or Jaccard distance.
10. The system of claim 7 , wherein the operations further comprise determining, based on merit attributes of training datasets, the first weight and the second weight using linear regression.
11. The system of claim 7 , wherein determining the merit attribute of the candidate dataset comprises:
determining a weighted sum of the number of the first unique candidate entities in the candidate dataset and the number of the second unique candidate entities in the candidate dataset,
wherein each of the first unique candidate entities is associated with the first weight; and
wherein each of the second unique candidate entities is associated with the second weight.
12. A system comprising:
means for receiving (i) a reference dataset identifying first entities associated with first features that include a baseline feature of a target population and (ii) a candidate dataset identifying second entities associated with second features;
means for identifying, in the candidate dataset, first unique candidate entities that are absent from the reference dataset and that are associated with the baseline feature in the candidate dataset;
means for forming, in a multi-dimensional space and based on a subset of the first features lacking the baseline feature, a cluster of data points representing the first entities, wherein forming the cluster of data points representing the first entities comprises:
generating an aggregated matrix that identifies the first entities, the first features, the second entities, and the second features;
estimating missing values in the aggregated matrix using spectral regularization; and
mapping, based on the aggregated matrix, the first entities to the data points representing the first entities, wherein coordinates of each data point in the data points representing the first entities are determined based on corresponding values in the aggregated matrix;
means for mapping a subset of the second entities that are absent from the reference dataset and that are not in the first unique candidate entities to additional data points, respectively in the multi-dimensional space, wherein mapping the subset of the second entities to the additional data points in the multi-dimensional space comprises mapping each second entity in the subset of the second entities to a respective additional data point in the multi-dimensional space based on corresponding values in the aggregated matrix;
means for identifying, from the subset of the second entities, second unique candidate entities corresponding to a subset of the additional data points within a threshold distance of the cluster;
means for determining a merit attribute of the candidate dataset based on a first weight for each first unique candidate entity, a second weight for each second unique candidate entity, a number of the first unique candidate entities in the candidate dataset, and a number of the second unique candidate entities in the candidate dataset; and
means for selecting the candidate dataset as input data for a target software application based on the merit attribute of the candidate dataset being greater than a threshold value.
13. The system of claim 12 , wherein the means for identifying the second unique candidate entities comprise:
means for determining, in the multi-dimensional space, a centroid of the cluster of the data points representing the first entities;
means for determining, in the multi-dimensional space, an average reference distance between each data point in the data points representing the first entities and the centroid of the cluster;
means for determining the threshold distance based on the average reference distance; and
means for determining, in the multi-dimensional space, a distance between a respective additional data point and the centroid of the cluster.
14. The system of claim 13 , wherein the distance between the respective additional data point and the centroid of the cluster includes a Pearson correlation distance, Euclidean distance, cosine distance, or Jaccard distance.
15. The system of claim 12 , wherein the means for determining the merit attribute of the candidate dataset comprise:
means for determining a weighted sum of the number of the first unique candidate entities in the candidate dataset and the number of the second unique candidate entities in the candidate dataset, wherein each of the first unique candidate entities is associated with the first weight; and
wherein each of the second unique candidate entities is associated with the second weight.