[ method should accept logical vectors / bool tensors

[t-kalinowski]

dispatch to tf.boolean_mask

[mohamed-180]

Does this will help :

if (is_tensor(dots[[1]]))
  if (dots[[1]]$dtype$is_bool) 
    tf$boolean_mask(x,dots[[1]])
   
  # ToDo make slice with tensors one based    

x <- tf$constant(1:6)
y <- tf$constant(c(T,T,T,F,F,T))
x[y]

tf.Tensor([1 2 3 6], shape=(4), dtype=int32)

[t-kalinowski]

Something like that would work I think. We'll also need it to work well for nd-arrays (not just 1d), nice error messages if a user tries to mix-n-match subsetting styles, and have a full test suite to go along with it.