Discussion: methods vs. function in R and Python about torch

[dfalbel]

IMHO we should closely follow python here. Ie if a method exists in Python it should also exist in R.

x$t() exists in R the side:

> x <- torch_randn(10, 5)
> x$t()
torch_tensor 
 1.5990 -1.2676  1.4600 -0.6393 -0.8593  1.0309  0.4477  1.0159  0.0463  0.0362
 2.1157  0.7238  0.0673  0.7179  0.1117 -1.0217 -2.2464  0.0867  1.2638 -0.7934
-1.2125 -1.0230 -0.0952 -1.1130  0.2201 -0.3773  0.5066 -0.8243 -0.3650 -0.2140
-0.2426  0.0644  0.2774 -0.9705  1.0606 -1.2880  0.8776 -0.4154  0.1012 -1.9847
-0.6819 -0.0620 -1.3663 -1.2802  0.7495 -0.3198 -1.4872 -0.9426  1.0952  0.2087
[ CPUFloatType{5,10} ]

clone is a special case, because it was not possible to add a clone method to an R6 class.
so we renamed it to copy. We should be able to call it clone again, since we are no longer using R6 classes for tensors.

[skeydan]

Oh, I completely agree! Perfect. Yes, then it'd be nice to have a clone as well!

Why I thought there was no x$t():

t$mm_(t$t())
Error: attempt to apply non-function

(but tbh I didn't investigate)

[dfalbel]

I think in this case, mm_ is the one that doesn't exist.

[skeydan]

oh, right - it really does not exist (in Python) :-)

Closing this and opening a separate one for clone :-)