Tensorflow在双向LSTM中并行运行cell_fw和cell_bw

Question

我正在尝试使用 tf while循环在双向 lstm 中运行 cell_fw 和 cell_bw body函数使用范围的不同参数（是python字符串）并行调用 lstm 函数。

#varaible declarations
....
....

lstm_scopes =['lstm_1','lstm_2']
def cond(act_outs_tensor, _it):
    # run the function twice
    return tf.less(_it, tf.constant(2))

def body(act_outs_tensor, _it):

    sess = tf.InteractiveSession()

    # get integer value of _it to get the req scope to pass 
    #  to pass to the function call
    idx = _it.eval() # throws an error saying items inside while are not fetchable
    lstm_scope = lstm_scopes[idx]
    act_outs_tensor = act_outs_tensor.write(_it,
                                              lstm(hidden_size_1,
                                                     dim, _input,
                                                     lstm_scope))
    return _it + 1, act_outs_tensor

_,act_outs_tensor = tf.while_loop(cond, body, [act_outs_tensor,_it],parallel_iterations=2)

上面的代码失败，因为“在TensorFlow条件的分支中创建的任何op或TensorFlow循环的主体都标记为“不可提取”，以防止发生各种编程错误” source。是否有任何变通办法可以在每次运行body函数时更改作用域的值？

谢谢