Construction method:
tf.placeholder(
dtype,
shape=None,
name=None
)Function description:
Insert a tensor (Tensor) placeholder to be initialized (it can be understood as opening up a space and assigning a specific value during execution )
Parameter Description:
dtype: The element type of the filled tensor (Tensor).
shape: The shape of the filled Tensor (optional parameter). If you do not specify a tensor (Tensor) shape, you can fill the tensor (Tensor) with any shape.
name: Provide a name for the operation (optional parameter).
return value:
A Tensor. Assignment must be done using a handle, but it cannot be evaluated directly.
Example:
Randomly generate a 1024*1024 matrix and find its square.
x = tf.placeholder(tf.float32, shape=(1024, 1024))
y = tf.matmul(x, x)
with tf.Session() as sess:
print(sess.run(y)) # ERROR: 由于x没有分配具体的数据所以该句将报错，注释掉后运行下面的语句即可
rand_array = np.random.rand(1024, 1024)
print(sess.run(y, feed_dict={x: rand_array})) # feed_dict以字典的方式将生成的rand_array填充进x中