Entropic clustering of objects

A semiconductor package apparatus, comprising:

• one or more substrates;
• and
• logic coupled to the one or more substrates, wherein the logic is at least partly implemented in one or more of configurable logic or fixed-functionality hardware logic, the logic coupled to the one or more substrates to:
• map a collection of data into two or more graph representations of the data based on a configurable set of rules that one of preserves or enhances relationships or properties of the data, wherein the collection of data is to include one or more of first data that has incomplete attributes for a feature, or second and third data that are associated with a same feature, wherein the second data is to include a first number of attributes for the same feature and the third data is to include a second number of attributes for the same feature, and wherein the first number is to be different from the second number,
• organize the two or more graph representations into a plurality of clusters of data based on graph information entropy and one or more parameters,
• select two or more clusters of data of the plurality of clusters of data, wherein the two or more clusters of data are associated with first graph information entropy per node,
• merge the two or more clusters of data to generate a merged cluster, wherein the merged cluster is associated with second graph information entropy per node,
• determine whether to accept the merged cluster based on a transition probability determined based on a graph information entropy of merging per node, wherein the graph information entropy of merging per node is to indicate whether the second graph information entropy per node is greater than the first graph information entropy per node, wherein the transition probability is to correspond to a probability that the merged cluster is accepted when the graph information entropy of merging per node indicates that the second graph information entropy per node is greater than the first graph information entropy per node,
• if the merged cluster is not accepted, select the two or more clusters and discard the merged cluster, and
• if the merged cluster is accepted, select the merged cluster and discard the two or more clusters.

A method of unsupervised machine learning, comprising:

• mapping a collection of data into two or more graph representations of the data based on a configurable set of rules that one of preserves or enhances relationships or properties of the data, wherein the collection of data includes one or more of first data that has incomplete attributes for a feature, or second and third data that are associated with a same feature, wherein the second data includes a first number of attributes for the same feature and the third data includes a second number of attributes for the same feature, and wherein the first number is different from the second number;
• organizing the two or more graph representations into a plurality of clusters of data based on graph information entropy and one or more parameters;
• selecting two or more clusters of data of the plurality of clusters of data, wherein the two or more clusters of data are associated with first graph information entropy per node;
• merging the two or more clusters of data to generate a merged cluster, wherein the merged cluster is associated with second graph information entropy per node;
• determining whether to accept the merged cluster based on a transition probability determined based on a graph information entropy of merging per node, wherein the graph information entropy of merging per node indicates whether the second graph information entropy per node is greater than the first graph information entropy per node, wherein the transition probability corresponds to a probability that the merged cluster is accepted when the graph information entropy of merging per node indicates that the second graph information entropy per node is greater than the first graph information entropy per node;
• if the merged cluster is not accepted, selecting the two or more clusters and discarding the merged cluster;
• and
• if the merged cluster is accepted, selecting the merged cluster and discarding the two or more clusters.

At least one non-transitory computer readable storage medium, comprising a set of instructions, which when executed by a computing device, cause the computing device to:

• map a collection of data into two or more graph representations of the data based on a configurable set of rules that one of preserves or enhances relationships or properties of the data, wherein the collection of data is to include one or more of first data that has incomplete attributes for a feature, or second and third data that are associated with a same feature, wherein the second data is to include a first number of attributes for the same feature and the third data is to include a second number of attributes for the same feature, and wherein the first number is to be different from the second number;
• organize the two or more graph representations into a plurality of clusters of data based on graph information entropy and one or more parameters;
• select two or more clusters of data of the plurality of clusters of data, wherein the two or more clusters of data are associated with first graph information entropy per node;
• merge the two or more clusters of data to generate a merged cluster, wherein the merged cluster is associated with second graph information entropy per node;
• determine whether to accept the merged cluster based on a transition probability determined based on a graph information entropy of merging per node, wherein the graph information entropy of merging per node is to indicate whether the second graph information entropy per node is greater than the first graph information entropy per node, wherein the transition probability is to correspond to a probability that the merged cluster is accepted when the graph information entropy of merging per node indicates that the second graph information entropy per node is greater than the first graph information entropy per node;
• if the merged cluster is not accepted, select the two or more clusters and discard the merged cluster;
• and
• if the merged cluster is accepted, select the merged cluster and discard the two or more clusters.