During implementation of different layers I noticed that ReduceMean divides by the full size of the matrix instead of only the reduced index sizes as expected by different implementations implemented R, Numpy, Matlab, etc.
Please have a look at the tests as an example.
Please feel free to adjust the coding style, optimize the implementation of the PR, mainly it was for a topic to discussion.
During implementation of different layers I noticed that ReduceMean divides by the full size of the matrix instead of only the reduced index sizes as expected by different implementations implemented R, Numpy, Matlab, etc.
Please have a look at the tests as an example.
Please feel free to adjust the coding style, optimize the implementation of the PR, mainly it was for a topic to discussion.
Thanks for the time and attention