## Description
MXNet has no `randint` implementation and integer sampling can be both limiting 
and confusing to developers:
* If using random_uniform with 'float32', the range is limited to 16.6M, beyond 
with there is no one to one mapping between float and int values.
* My tests show that random_uniform with 'float64' seem to address the issue, 
but this is not obvious to most developers and it may be extra unnecessary 
precision to achieve the necessary accuracy.
* It isn't very well known how to do proper random integer sampling using an 
underlying float sampling, i.e.:
```
randint = cast(clip(random.uniform(low, high, shape), low, high-1), 'int32')
```  

A `randint` operator implementation will solve all above issues.


[ Full content available at: 
https://github.com/apache/incubator-mxnet/issues/12478 ]
This message was relayed via gitbox.apache.org for [email protected]

Reply via email to