Hi cuda,
Thank you for a quick answer!
It can be useful for recommender system algorithms.
Such algorithms deal with initialization of input matices (which are products of SVD) having huge data array (such as 500 000 x 500 000) of sparse data.
The resulting SVD matrices are 3*500Kx500K (Dense). As a result we cannot fit all these data volume in the memory and therefore we cannot use SVD from Math.Net.
Can you please say if I miss something in my undestanding of memory usage for resulting SVD matrices?
If I am correct - I'll write a feature request with explanations why it can be useful.
Have a nice day!
Best regards, Alex.
Thank you for a quick answer!
It can be useful for recommender system algorithms.
Such algorithms deal with initialization of input matices (which are products of SVD) having huge data array (such as 500 000 x 500 000) of sparse data.
The resulting SVD matrices are 3*500Kx500K (Dense). As a result we cannot fit all these data volume in the memory and therefore we cannot use SVD from Math.Net.
Can you please say if I miss something in my undestanding of memory usage for resulting SVD matrices?
If I am correct - I'll write a feature request with explanations why it can be useful.
Have a nice day!
Best regards, Alex.
