The output of K.dropout is not as expected!

[Imorton-zd]

For example:

x = numpy.array[[1,2,3],[4,5,6]]
x = K.variable(value = x, dtype = 'float64')
out = K.dropout(x, 0.4, seed=1334)
K.eval(out)

The results:
array([1.6666666, 3.3333332, 4.9999998], [0. , 8.33333 , 0. ])

But, in my opinion, the dropout operator only sets entries in x to zero at random, the results should be like: array([1, 2, 3], [0. , 5 , 0. ])
why are other elements also modified?
If I want to implement the function like my description, which method should I use? Any opinions or suggestions would be appreciated!

[unrealwill]

Hello,

Because Keras use inverse scaling for dropout.

See #3305
and #5357

for more information.

[Imorton-zd]

Thanks for your reply.