Tensorflow: `tf.reshape((), (0))` works fine in eager mode but ValueError in Graph mode-CodePudding

As the title, the function tf.reshape((), (0)) works perfectly fine in eager mode. But when I use it in Graph mode, it returns:
ValueError: Shape must be rank 1 but is rank 0 for '{{node Reshape}} = Reshape[T=DT_FLOAT, Tshape=DT_INT32](Reshape/tensor, Reshape/shape)' with input shapes: [0], [].

Can anyone help me with the work-around of this function please. You can reproduce this error here. I want to recreate this piece of code (initialization of mems) in Tensorflow's graph mode. Thanks in advance!!

Edit: Since I want to clarify my use case, I want to convert this PyTorch torch.empty(0) function to Tensorflow (work fine in both eager mode and graph mode).

CodePudding user response：

Might be related to this bug. Try something like this:

@tf.function
def test_graph():
    x = tf.reshape((), (0, ))
    return x

b = test_graph()
b
#<tf.Tensor: shape=(0,), dtype=float32, numpy=array([], dtype=float32)>