TF基本概念
Graph 表示计算任务
Node 可以是Operation也可以是数据存储容器
在session的context中执行graph
使用tensor表示数据
通过variable维护状态
使用feed和fetch 为任意操作赋值(arbitrary operation)或者从中获取数据
Tensor 类似于numpy 中的数组
3# a rank 0 tensor; this is a scalar withshape []
[1. ,2., 3.] # a rank 1tensor; this is a vector with shape [3]
[[1., 2., 3.], [4., 5., 6.]] # a rank 2tensor; a matrix with shape [2, 3]
[[[1., 2., 3.]], [[7., 8., 9.]]] # a rank 3tensor with shape [2, 1, 3]
Computational Graph包括两步
1、 Building the computational graph.
将TF操作转换成Graph nodes的形式,每个node包括input Tensor 和 output Tensor;constant node 只有固定输入没有输出
Eg:
import tensorflow as tf
node1 = tf.constant(3.0, tf.float32)
node2 = tf.constant(4.0) # also tf.float32 implicitly
print node1, node2
[out]
Tensor("Const:0", shape=TensorShape([]), dtype=float32)Tensor("Const_1:0", shape=TensorShape([]), dtype=float32)
sess = tf.Session()
print sess.run([node1,node2])
[out]
[3.0, 4.0]
将上述两个node 相加产生新的node 并输出computation graph
node3 = tf.add(node1, node2)
print "node3: ", node3
print "sess.run(node3):",sess.run(node3)
[out]
node3: Tensor("Add:0", shape=TensorShape([]), dtype=float32)
sess.run(node3): 7.0
Placeholders可以不需要在定义的时候赋值,可以随后赋值
EG:
IN:
a = tf.placeholder(tf.float32)
b = tf.placeholder(tf.float32)
adder_node = a + b # + provides a shortcut for tf.add(a, b)
print sess.run(adder_node, {a: 3, b:4.5})
print sess.run(adder_node, {a: [1,3], b:[2, 4]})
OUT:
7.5
[ 3. 7.]
在此基础上在加一个graph
IN:
add_add_triple = adder_node * 3.
print sess.run(add_add_triple,{a:3,b:4.5})
OUT:
22.5
2、Running thecomputational graph.
为了计算node 值(3.0)(4.0)必须使用Session
Variables 可以将trainableparameters 加到graph中
构造Variable需要类型和初值
IN:
W = tf.Variable([.3], tf.float32)
b = tf.Variable([-.3], tf.float32)
x = tf.placeholder(tf.float32)
linear_model = W * x +b
#variable initialize needs call a specialoperation
init = tf.initialize_all_variables()
sess.run(init)
print sess.run(linear_model,{x:[1,2,3,4]})
OUT:
[ 0. 0.30000001 0.60000002 0.90000004]
计算loss func
IN:
############ y and calculate loss function
y = tf.placeholder(tf.float32)
squared_deltas = tf.square(linear_model -y)
loss = tf.reduce_sum(squared_deltas)
printsess.run(loss,{x:[1,2,3,5],y:[0,-1,-2,-3]})
OUT:
26.09
手动调参 tf.assign
####adjust the parametres by hand
IN:
fixW = tf.assign(W,[-1.])
fixb = tf.assign(b,[1.])
sess.run([fixW, fixb])
printsess.run(loss,{x:[1,2,3,4],y:[0,-1,-2,-3]})
OUT:
0.0
自动调参数gradient decent
IN:
#gradientDescent optimeze the parameter
optimizer =tf.train.GradientDescentOptimizer(0.01)
train = optimizer.minimize(loss)
sess.run(init)
for i in range(1000):
sess.run(train,{x:[1,2,3,4],y:[0,-1,-2,-3]})
print sess.run([W,b])
[array([-0.9999969], dtype=float32),array([ 0.99999082], dtype=float32)]