MyYaYa commented on issue #13001: Feature request: numpy.cumsum
URL:
https://github.com/apache/incubator-mxnet/issues/13001#issuecomment-451343665
> I find a elegant method to implement cumsum. Hope this will be useful.
>
> ```
> step = lambda data, state : (data + state[0], state
MyYaYa commented on issue #13001: Feature request: numpy.cumsum
URL:
https://github.com/apache/incubator-mxnet/issues/13001#issuecomment-447198557
@tsuberim yeah, as I know, the symbol api in MXNet support
auto-differentiation, so you can try the mx.sym.contrib.foreach(), maybe that
will
MyYaYa commented on issue #13001: Feature request: numpy.cumsum
URL:
https://github.com/apache/incubator-mxnet/issues/13001#issuecomment-446866056
I find a elegant method to implement cumsum. Hope this will be useful.
```
step = lambda data, state : (data + state[0], state[0] + data)
MyYaYa commented on issue #13001: Feature request: numpy.cumsum
URL:
https://github.com/apache/incubator-mxnet/issues/13001#issuecomment-446865050
I find a elegant method to implement cumsum. Hope this will be useful.
`
step = lambda data, state : (data + state[0], state[0] + data
MyYaYa commented on issue #13001: Feature request: numpy.cumsum
URL:
https://github.com/apache/incubator-mxnet/issues/13001#issuecomment-446864884
`step = lambda data, state : (data + state[0], state[0] + data)
data = mx.nd.ones((10, 10))
state = mx.nd.zeros(10)
outs, state = mx.nd
MyYaYa commented on issue #13001: Feature request: numpy.cumsum
URL:
https://github.com/apache/incubator-mxnet/issues/13001#issuecomment-446489801
Yeah, if ndarray supports cumsum operation, some custom metrics(i.e. mAP)
will benefit from that.