[GitHub] jermainewang commented on issue #7082: Sparse Tensor: request for reviews

[gi...@git.apache.org]

jermainewang commented on issue #7082: Sparse Tensor: request for reviews
URL: https://github.com/apache/incubator-mxnet/pull/7082#issuecomment-319390628
 
 
   The storage format can be formalized as a vector. Each element is a pair contains "dim" and "dense/sparse".
   For example:
   * Row-major equals to [<0, dense>, <1, dense>]
   * Col-major equals to [<1, dense>, <0, dense>]
   * CSR equals to [<0, dense>, <1, sparse>]
   * CSC equals to [<1, dense>, <0, sparse>]
   
   If having more tensor dimensions (e.g. convolution),
   * BCSR equals to [<0, dense>, <1, sparse>, <2, dense>, <3, dense>]
   
   I found most variants of current sparse tensor formats can be formalized as this. The paper talks about coordinate, ellpack and dia formats in the future work, which they believe can also be supported.
   
   In terms of implementation, we could change the enum to the vector above, define some common formats (like CSR, CSC), and define the "==" operator. This should simulate the behavior of enum type.
 
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services