- constants (no inputs, outputs a value it stores internally)
tf.constant(3.0) - operations (inputs to outputs)
tf.add(node1, node2)ornode1 + node2 - (Variables and placeholders are operations too. For variables, an 'assign' op might also be created for assigning the initial value)
3 # a rank 0 tensor; this is a scalar with shape []
[1. ,2., 3.] # a rank 1 tensor; this is a vector with shape [3]
[[1., 2., 3.], [4., 5., 6.]] # a rank 2 tensor; a matrix with shape [2, 3]
[[[1., 2., 3.]], [[7., 8., 9.]]] # a rank 3 tensor with shape [2, 1, 3]-
A computational graph is a series of TF operations arranged into a graph of nodes.
-
Each node takes zero or more tensors as inputs and produces a tensor as an output.
-
To actually evaluate the nodes, we must run the computational graph within a session:
sess = tf.Session(); sess.run([node1, node2]) -
A graph can be parameterized to accept external inputs, known as placeholders. A placeholder is a promise to provide a value later.
-
a = tf.placeholder(tf.float32),adder_node = a + b -
To specify the placeholder values (tensors) when evaluating the graph, pass a
feed_dictparameter:sess.run(adder_node, {a: [1,3], b: [2, 4]}) -
Variables allow us to add trainable parameters to a graph. They are constructed with a type and initial value:
W = tf.Variable([.3], dtype=tf.float32)
b = tf.Variable([-.3], dtype=tf.float32)
x = tf.placeholder(tf.float32)
linear_model = W * x + b- To initialize all the variables in a TensorFlow program:
init = tf.global_variables_initializer(); sess.run(init)
- Example of evaluating model's loss:
y = tf.placeholder(tf.float32)
squared_deltas = tf.square(linear_model - y)
loss = tf.reduce_sum(squared_deltas)
print(sess.run(loss, {x:[1,2,3,4], y:[0,-1,-2,-3]}))- Variable values can be changed with e.g.
tf.assignoperation:fixW = tf.assign(W, [-1.]); sess.run(fixW) - But usually we'll train with an optimizer to minimize the loss by changing the variables' values:
optimizer = tf.train.GradientDescentOptimizer(0.01)
train_op = optimizer.minimize(loss)
sess.run(init) # reset values to incorrect defaults.
for i in range(1000):
sess.run(train, {x:[1,2,3,4], y:[0,-1,-2,-3]})
print(sess.run([W, b]))