[GitHub] [incubator-mxnet] hetong007 opened issue #12319: Feature Request: support several statistical operators in NDArray

["hetong007 (GitHub)" <gi...@apache.org>]

## Description

Median, variance and standard deviation are basic, useful and descriptive statistics of an array of numbers.

I propose to add these basic operators for NDArray.

@haojin2 do you have any comments on this?


[ Full content available at: https://github.com/apache/incubator-mxnet/issues/12319 ]
This message was relayed via gitbox.apache.org for devnull@infra.apache.org

[GitHub] [incubator-mxnet] haojin2 commented on issue #12319: Feature Request: support several statistical operators in NDArray

["haojin2 (GitHub)" <gi...@apache.org>]

Are we looking to match the same behaviors of the Numpy counterparts?

[ Full content available at: https://github.com/apache/incubator-mxnet/issues/12319 ]
This message was relayed via gitbox.apache.org for devnull@infra.apache.org

[GitHub] [incubator-mxnet] lupesko commented on issue #12319: Feature Request: support several statistical operators in NDArray

["lupesko (GitHub)" <gi...@apache.org>]

@hetong007 - are there examples of models or use cases where these are useful? would help me to see if I can prioritize this on my team in Amazon. Thanks!  

[ Full content available at: https://github.com/apache/incubator-mxnet/issues/12319 ]
This message was relayed via gitbox.apache.org for devnull@infra.apache.org

[GitHub] [incubator-mxnet] hetong007 commented on issue #12319: Feature Request: support several statistical operators in NDArray

["hetong007 (GitHub)" <gi...@apache.org>]

@haojin2 I'd say as long as numpy counterparts are mathematically correct. 

[ Full content available at: https://github.com/apache/incubator-mxnet/issues/12319 ]
This message was relayed via gitbox.apache.org for devnull@infra.apache.org

[GitHub] [incubator-mxnet] hetong007 commented on issue #12319: Feature Request: support several statistical operators in NDArray

["hetong007 (GitHub)" <gi...@apache.org>]

@lupesko , thanks for your response. 

My use case was in network quantization, i.e. train the network so that the weights are quantized from float32/float16 into low-bit values and some methods takes the median on the array objects. 

For variance and standard deviation, I would like to have them to descriptively analyze the initialization and distribution of the weights.

[ Full content available at: https://github.com/apache/incubator-mxnet/issues/12319 ]
This message was relayed via gitbox.apache.org for devnull@infra.apache.org