Javascript learned or should xml students a better understanding of the DOM model,
DOM model, all things are seen as Node element,
but there are different sub-Node element earth elements, such as document, element, attribute, etc.,
and is very similar to DOM , tensorflow all elements called grarh elements, together constituting a gragh,
and specifically, there are the following several basic elements:
tensor (tensor):
as already said.
Variable:
also known weight (weight), parameter (w), there are between two nodes connected to such a parameter,
the future of training is to find the right parameters.
general directly to all the parameters between the two layers is written inside a matrix,
the number of rows to the starting point node layer number, the number of columns for the end layer nodes.
placeholder:
a lot of people called placeholders, in fact, the input vector variables,
like function definitions, wrote parameter list parameter.
the main features of the number of rows in a statement here matrix future (Shape),
such as: X = tf.placeholder (tf.float32, Shape = (None, 2))
OP:
node acquires tensor (or not to acquire tensor), calculated tensor,
described between tensor computing relationship, it is the real structure of the network.
general computing section Such as:
A = tf.matmul (X, W1)
Y = tf.matmul (A, w2 of)
There are special types of nodes (I've seen):
1. loss function:
such as: Loss = tf.reduce_mean (tf.square (Y_-Y))
2. Reverse propagation:
train_step = tf.train.GradientDescentOptimizer (from 0.001) .minimize (Loss)
train_step = tf.train.MomentumOptimizer (0.0001,0.99) .minimize (Loss)
train_step = tf.train.AdamOptimizer (from 0.0001) .minimize (Loss)
(in practice only one is enough)
3. variables initialization:
Yes, you read right, variable initialization even be considered graph elements,
and then they have to initialize before you can "run" behind the node
init_op = tf.global_variables_initializer ()
sess.run (init_op)
4. counter:
current level I see I do not know, do not write.
