TensorArray

由tensor构成的array，可以动态地向里面加入tensor，所以常用来与loop等flow control op一起用

构造
tf.TensorArray(size=1, dynamic_size=True)
size代表初始长度，dynamic_size为true表示可以动态的向里面添加tensor

写入Tensor
output_ta = output_ta.write(index=0,value=tf.constant(0,shape=[3,2]))
将tf.constant(0,shape=[3,2])这个tensor写到index为0的位置
该函数返回 A new TensorArray object with flow that ensures the write occurs. 所以之后的操作要使用这个返回的TensorArray object

读取Tensor
t = output_ta.read(index = 0)
读取index为0的tensor
返回的是tensor

将TensorArray转换成Tensor
t = output_ta.stack(): stack on first dimension
t = output_ta.gather(indices = [0,2])
t = output_ta.concat()

将Tensor转化成TensorArray
output_ta = output_ta.unstack(value = t)
output_ta = output_ta.scatter(indices = [0,2], value= t):Scatter the values of a `Tensor` in specific indices of a `TensorArray`
output_ta = output_ta.split(value = t, lengths = [3])