Tensor

Tensor has two key properties:
1. Shape
2. Dtype
Dtype is always known. 
The shape might be only partially known. Most operations produce tensors of fully-known shapes if the shapes of their inputs are also fully known. In some cases it's only possible to find the shape of a tensor at graph execution time.
The value of a tensor is immutable

Rank
1. rank = 0: scalar, so to produce a scalar specify shape = []
    Tensor with shape = [1] is not a scalar
2. r = tf.rank(tensor), r is a tensor need to be run by session
3. string tensor是一个scalar tensor
mammal = tf.Variable("Elephant", tf.string)
ignition = tf.Variable(451, tf.int16)
floating = tf.Variable(3.14159265359, tf.float64)

Shape
There are two ways of accessing the shape of a tf.Tensor
1. This can be done by reading the shape property of a tf.Tensor object. This method returns a TensorShape object (only partiallty known shape): a.shape[0].value，可以直接得到数值
2. The tf.shape operation will represent the fully-defined shape of another tf.Tensor: 返回的是一个tensor，可以在运行时确定tensor的形状

Referring to tf.Tensor slices
1. specify n indices for rank = n tensor
  For a rank 0 tensor (a scalar), no indices are necessary, since it is already a single number
  For rank 1: my_scalar = my_vector[2], Note that the index passed inside the [] can itself be a scalar tf.Tensor, if you want to dynamically choose an element from the vector.
  For rank 2: my_scalar = my_matrix[1, 2] (get a scalar); my_column_vector = my_matrix[:, 3],维度
  slice得到的tensor会自动降维