[R] Take care of the input type of mx.nd.array in the R's interface

[hetong007]

In the R's interface, I suggest to wrap up the as.array operation in mx.nd.array, so users can simply write

mat = mx.nd.array(1:10, mx.cpu(0))
mat = mx.nd.array(matrix(1:4,2,2), mx.cpu(0))

instead of

x = as.array(c(1,2,3))
mat = mx.nd.array(x, mx.cpu(0))

in the current demo like https://github.com/dmlc/mxnet/blob/master/R-package/demo/basic_ndarray.R.

[tqchen]

@thirdwing I think we can do a type version check in rcpp side?

[tqchen]

I now started to think that we should do such change on R's side as rewrap the mx.nd.array is fairly easy from R's side. Let us move mx.nd.array to mx.nd.internal.array and create the wrapper as tong suggested.

@hetong007 Can you do that? You only need to change https://github.com/dmlc/mxnet/blob/master/R-package/src/ndarray.cc#L387 On cpp side to mx.nd.internal.array

[hetong007]

I think so. I think I will modify things tmr, e.g. travis ci, this wrapper and the symbolic ones.

[tqchen]

Great. thanks tong