R6类方法中的R张量流计算返回错误

Question

在R中我有一个带有预测方法的R6类。当调用预测方法时，它应该运行张量流计算，这里乘以输入和权重。

这是代码：

NeuralNetworkTensorflow = R6::R6Class("NeuralNetworkTensorflow",
  public = list(
    inputs = NULL,
    weights = NULL,
    Q = NULL,
    sess = NULL,

    initialize = function() {
      tf$reset_default_graph()
      self$inputs = tf$placeholder(tf$float32, shape(1L, 5L))
      self$weights = tf$Variable(tf$random_uniform(shape(5L, 4L), 0, 0.01)) 
      self$Q = tf$matmul(self$inputs, self$weights)

      self$sess = tf$Session()
      self$sess$run(tf$global_variables_initializer())
    },

    predict = function(inputs) {
      self$sess$run(self$Q, feed_dict = dict(inputs = inputs))
    }
  )
)

my_nn = NeuralNetworkTensorflow$new()
my_nn$predict(matrix(c(0, 1, 0, 0, 1), ncol = 5))

它返回以下错误：

 Error in py_call_impl(callable, dots$args, dots$keywords) : 
  SystemError: <built-in function TF_NewBuffer> returned a result with an error set

我不确定，这里出了什么问题。有人可以帮忙吗？