Scalable topological data analysis

1. A method comprising:

receiving first data associated with data points;

receiving a lens function selection, a metric function selection, and a resolution function, the metric function identified by the metric function selection being capable of performing functions on data as matrix functions;

mapping second data based on the first data to a reference space by utilizing matrix vector multiplication for application of selected lens function on second data based on the first data to map the second data to the reference space, the results of the matrix vector multiplication being stored in temporary memory enabling reduction of space complexity of memory;

generating cover of reference space including the second data;

clustering second data in cover based on the selected metric function to determine each node of a plurality of nodes, each of the nodes of the plurality of nodes comprising members representative of at least one subset of the data points; and

generating a visualization comprising the plurality of nodes and a plurality of edges wherein each of the edges of the plurality of edges connects nodes with shared members.

2. The method of claim 1 , wherein the metric function is a cosine distance function representable in a matrix form:

D cos=11 T −diag(diag( XX T )) −1/2 XX T diag(diag( XX T )) −1/2

where again X is a matrix that stores a dataset, “1” is a vector of all ones and “1 T ” is a transpose of the vectors of all ones.

3. The method of claim 1 , wherein the metric function is a correlation metric function representable in a matrix form:

D =11 T −diag(diag( YY T )) −1/2 YY T diag(diag( YY T )) −1/2

where Y=X−μ1 T .

4. The method of claim 1 , wherein the metric function is a Hamming metric function representable in a matrix form:

D hamming =X (11 T −X T )

5. The method of claim 1 , wherein the metric function is an L 2 metric function representable in a matrix form:

D∘D=− 2 XX T +diag( XX T )1 T +1diag( XX T ) T

6. The method of claim 1 , wherein the lens function is an L 1 centrality function wherein given a distance operator D based on the metric function, the L 1 centrality of each row in a data matrix is encoded in a vector D·1.

7. The method of claim 1 , wherein the lens function is an L 2 centrality function wherein given a distance operator D based on the metric function, the L 2 centrality of each row in a data matrix is encoded in an entry-wise square root of a vector D∘D·1, wherein D∘D is a Hadamard square.

8. The method of claim 1 , wherein the lens function is a gaussian density function and wherein given a distance operator D based on the metric function, an entry-wise function exp(−(·)) is performed on a Hadamard square of the distance operator to each entry.

9. The method of claim 1 , wherein the lens function is a PCA function wherein the first data is itself multiplied.

10. The method of claim 1 , wherein the lens function is a metric PCA function that utilizes a distance operator D based on the metric function.

11. The method of claim 1 , wherein the lens function is an MDS function that utilizes a Hadamard square of a distance operator D which is based on the metric function.

12. A non-transitory computer readable medium comprising instructions executable by a processor to perform a method, the method comprising:

receiving first data associated with data points;

receiving a lens function selection, a metric function selection, and a resolution function, the metric function identified by the metric function selection being capable of performing functions on data as matrix functions;

mapping second data based on the first data to a reference space by utilizing matrix vector multiplication for application of selected lens function on second data based on the first data to map the second data to the reference space, the results of the matrix vector multiplication being stored in temporary memory enabling reduction of space complexity of memory;

generating cover of reference space including the second data;

clustering second data in cover based on the selected metric function to determine each node of a plurality of nodes, each of the nodes of the plurality of nodes comprising members representative of at least one subset of the data points; and

generating a visualization comprising the plurality of nodes and a plurality of edges wherein each of the edges of the plurality of edges connects nodes with shared members.

13. The non-transitory computer readable medium of claim 12 , wherein the metric function is a cosine distance function representable in a matrix form:

D cos=11 T −diag(diag( XX T )) −1/2 XX T diag(diag( XX T )) −1/2

where again X is a matrix that stores a dataset, “1” is a vector of all ones and “1 T ” is a transpose of the vectors of all ones.

14. The non-transitory computer readable medium of claim 12 , wherein the metric function is a correlation metric function representable in a matrix form:

D= 11 T −diag(diag( YY T )) −1/2 YY T diag(diag( YY T )) −1/2

where Y=X−μ1 T .

15. The non-transitory computer readable medium of claim 12 , wherein the metric function is a Hamming metric function representable in a matrix form:

D hamming =X (11 T −X T )

16. The non-transitory computer readable medium of claim 12 , wherein the metric function is an L 2 metric function representable in a matrix form:

D∘D=− 2 XX T +diag( XX T )1 T +1diag( XX T ) T

17. The non-transitory computer readable medium of claim 12 , wherein the lens function is an L 1 centrality function wherein given a distance operator D based on the metric function, the L 1 centrality of each row in a data matrix is encoded in a vector D·1.

18. The non-transitory computer readable medium of claim 12 , wherein the lens function is an L 2 centrality function wherein given a distance operator D based on the metric function, the L 2 centrality of each row in a data matrix is encoded in an entry-wise square root of a vector D∘D·1, wherein D∘D is a Hadamard square.

19. The non-transitory computer readable medium of claim 12 , wherein the lens function is a gaussian density function and wherein given a distance operator D based on the metric function, an entry-wise function exp(−(·)) is performed on a Hadamard square of the distance operator to each entry.

20. The non-transitory computer readable medium of claim 12 , wherein the lens function is a PCA function wherein the first data is itself multiplied.

21. The non-transitory computer readable medium of claim 12 , wherein the lens function is a metric PCA function that utilizes a distance operator D based on the metric function.

22. The non-transitory computer readable medium of claim 12 , wherein the lens function is an MDS function that utilizes a Hadamard square of a distance operator D which is based on the metric function.

23. A system comprising:

one or more processors; and

memory containing instructions executable by the processor to:

receive first data associated with data points;

receive a lens function selection, a metric function selection, and a resolution function, the metric function identified by the metric function selection being capable of performing functions on data as matrix functions;

map second data based on the first data to a reference space by utilizing matrix vector multiplication for application of selected lens function on second data based on the first data to map the second data to the reference space, the results of the matrix vector multiplication being stored in temporary memory enabling reduction of space complexity of memory;

generate cover of reference space including the second data;

cluster second data in cover based on the selected metric function to determine each node of a plurality of nodes, each of the nodes of the plurality of nodes comprising members representative of at least one subset of the data points; and

generate a visualization comprising the plurality of nodes and a plurality of edges wherein each of the edges of the plurality of edges connects nodes with shared members.