1. Tensorflow Lecture
박 영 택
컴퓨터학부

2. tensorflow start # tensorflow start import tensorflow as tf
c1 = tf.constant(1)
c2 = tf.constant(2)
c3 = tf.mul(c1, c2)
with tf.Session() as sess:
print "c3 : ", sess.run(c3)
>> c3 : 2

3. tensorflow constants # tensroflow constants import tensorflow as tf
matrix1 = tf.constant([[3,3]])
matrix2 = tf.constant([[2],[2]])
product = tf.matmul(matrix1, matrix2)
sess = tf.Session()
print sess.run(matrix1)
with sess : #sess.run(t) = t.eval()
print matrix2.eval()
print product.eval()
>> [ [3 3] ]
[ [2]
[2] ]
[ [12] ]

4. tensorflow variables
# tensorflow variables
import tensorflow as tf
state = tf.Variable(0, name = "counter" )
print "state : ", state
init_op= tf.initialize_all_variables()
with tf.Session() as sess:
sess.run(init_op)
print "sess.run(state) : ", sess.run(state)
state.assign(1)
print "state.assign(1) : ", sess.run(state)
assign_op = state.assign(1)
sess.run(assign_op)
print "sess.run(state) after run(assign_op) : ", sess.run(state)
print state
>> state : <tensorflow.python.ops.variables.
Variable object at 0x1078e8390>
sess.run(state) : 0
state.assign(1) : 0
sess.run(state) after run(assign_op) : 1
<tensorflow.python.ops.variables.

5. a simple example >> state : 0
import tensorflow as tf
state = tf.Variable(0, name="counter")
one = tf.constant(1)
new_value = tf.add(state,one)
update = tf.assign(state, new_value)
init_op = tf.initialize_all_variables()
with tf.Session() as sess :
sess.run(init_op)
print "state : ", sess.run(state)
sess.run(update)
print "state after update : ", state
print "state after update : ", sess.run(state)
>> state : 0
state after update : <tensorflow.python.ops.variables.
Variable object at 0x107bc7490>
state after update : 1

6. representation of shape
import tensorflow as tf
a = tf.zeros([5,])
a1 = tf.zeros((5,))
b = tf.zeros([2,3])
c = tf.zeros((1,2,3))
with tf.Session() as sess:
print "a : ", a
print "sess.run(a) : ", sess.run(a)
print "sess.run(b) : ", sess.run(b)
print "sess.run(c) : ", sess.run(c)
print "shape of a :", a.get_shape()
print "shape of b :", b.get_shape()
print "shape of c :", c.get_shape()
>> a : Tensor("zeros_18:0", shape=TensorShape([Dimension(5)]), dtype=float32)
sess.run(a) : [ ]
sess.run(b) : [[ ] [ ]]
sess.run(c) : [[[ ] [ ]]]
shape of a : TensorShape([Dimension(5)])
shape of b : TensorShape([Dimension(2), Dimension(3)])
shape of c : TensorShape([Dimension(1), Dimension(2), Dimension(3)])

7. random uniform : shape, min, max
import tensorflow as tf
a = tf.random_uniform((2,3),-1,1)
b = tf.random_uniform((2,3), 90,100)
with tf.Session() as sess:
print "a : ", sess.run(a)
print "b : ", sess.run(b)
print "shape of a : ", a.get_shape()
>> a : [[ ] [ ]]
b : [[ ] [ ]]
shape of a : TensorShape([Dimension(2), Dimension(3)])

8. place holder #place holder import tensorflow as tf
#x = tf.placeholder("float", shape=(42, 4))
#x = tf.placeholder("float", shape=[42, 4])
x = tf.placeholder("float", (2, 3))
y = tf.zeros([2, 3], "float")
print(x.get_shape())
print(y.get_shape())
>> TensorShape([Dimension(2), Dimension(3)])
TensorShape([Dimension(2), Dimension(3)])

9. matrix add # matrix add import tensorflow as tf
x = tf.constant(([1,2], [3,4]))
y = tf.constant([[10,20], [30,40]])
z = tf.add(x, y)
with tf.Session() as sess:
print sess.run(z)
>> [[ ]
[ ]]

10. place holder example #place holder example import tensorflow as tf
x = tf.placeholder("float", (2, 3))
y = tf.zeros([2, 3], "float")
z = tf.add(x, y)
with tf.Session() as sess:
print sess.run(z, feed_dict={x: [[1,2,3], [4,5,6]]})
>> [[ ]
[ ]]

11. place holder example #place holder example import tensorflow as tf
x = tf.placeholder("float", (2, 3))
y = tf.placeholder("float", [2, 3])
z = tf.add(x, y)
with tf.Session() as sess:
print sess.run(z, feed_dict={x: [[1,2,3], [4,5,6]], y:[[10,20,30], [40,50,60]]})
>> [[ ]
[ ]]

12. a simple linear regression test
import numpy as np
import tensorflow as tf
b = tf.Variable(tf.zeros((3,))) # b shape = 1X3
W = tf.Variable(tf.random_uniform((4,3), -1, 1)) # W shape = 4 X 3
x = tf.placeholder(tf.float32, (None, 4)) # x shape = None X 4
h_i = tf.nn.relu(tf.matmul(x,W) + b)
c = tf.matmul(x,W)
init_op = tf.initialize_all_variables()
with tf.Session() as sess:
sess.run(init_op)
print "h_i : \n", sess.run(h_i, {x:np.random.rand(2,4)})
print "W: \n", sess.run(W)
print "c : matmul([1,1,1,1],W) \n", sess.run(c, {x: [[1,1,1,1]]})
>> h_i :
[ [ ]
[ ] ] W:
[ [ ]
[ ]
[ ]
[ ] ]
c : matmul([1,1,1,1],W)
[ [ ] ]