R/print.R

In tfautograph: Autograph R for 'Tensorflow'

ag_print <- function(x) {
  .NotYetImplemented()
  # Questioning if this is even a good idea to implement. It would only really
  # be useful in tensorflow v1 outside of a tf.function() context. Everywhere
  # else, tf$print() gets the job done with would the need to manage any
  # tf.control_dependencies() contexts.
  env <- parent.frame()
  x_expr <- substitute(x)
  sym_nms <- all.vars(x_expr)
  op <- tf$print(x)
  new_sym_vals <-
    with(tf$control_dependencies(list(op)),
         lapply(mget(sym_nms, env), function(val) tf$identity(val)))

  list2env(new_sym_vals, env)
  # what if sym was pulled from some parent env? Is this still the best thing to
  # do? At least this way works with ag_if(), but it'll break relatively
  # esoteric uses of local()
}