

In [2]:

    
from __future__ import print_function
import tensorflow as tf



In [3]:

    
# apt-get install graphviz
# pip install graphviz
from tfdot import tfdot

Basic



In [4]:

    
matrix1 = tf.constant([[3., 3.]])

matrix2 = tf.constant([[2.],[2.]])

product = tf.matmul(matrix1, matrix2)



In [5]:

    
tfdot()









    Out[5]:

Session



In [6]:

    
sess = tf.Session()
print(sess.run(product))
sess.close()









    



[[ 12.]]

Context manager



In [8]:

    
with tf.Session() as sess:
    print(sess.run(product))









    



[[ 12.]]

Device context



In [10]:

    
with tf.Session(config=tf.ConfigProto(log_device_placement=True)) as sess:
    with tf.device("/cpu:0"):
        m1 = tf.constant([[3., 3.]])
        m2 = tf.constant([[2.],[2.]])
        pd = tf.matmul(m1, m2)
        print(sess.run(pd))









    



[[ 12.]]

Interactive session



In [11]:

    
sess = tf.InteractiveSession()



In [12]:

    
product.eval()









    Out[12]:





array([[ 12.]], dtype=float32)



In [13]:

    
pd.eval()









    Out[13]:





array([[ 12.]], dtype=float32)

Variable



In [14]:

    
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()



In [15]:

    
tfdot()









    Out[15]:



In [17]:

    
init_op.run()
# state.run() # variables do not run
print(state.eval())



In [18]:

    
for _ in range(300):
    update.eval()



In [19]:

    
state.eval()









    Out[19]:





300



In [20]:

    
sess.run([update]*10)









    Out[20]:





[301, 301, 301, 301, 301, 301, 301, 301, 301, 301]

Initialize from another variable



In [21]:

    
# will back to here later
# tf.ops.reset_default_graph()
# sess.close()
# sess = tf.InteractiveSession()



In [22]:

    
weights = tf.Variable(tf.random_normal([784, 200], stddev=0.35), name='weights')
w2 = tf.Variable(weights.initialized_value(), name ='w2')
w_twice = tf.Variable(weights.initialized_value()*0.2, name="w_twice")



In [23]:

    
tfdot()









    Out[23]:



In [24]:

    
init_op = tf.initialize_all_variables()
init_op.run()



In [25]:

    
weights.eval()









    Out[25]:





array([[ 0.04481283, -0.62056214, -0.69336498, ..., -0.30848733,
        -0.34599236, -0.42386416],
       [-0.78550404, -0.10613412,  0.09037922, ...,  0.66008407,
         0.57086742, -0.1224874 ],
       [-0.61029941,  0.35488346, -0.10510538, ..., -0.10046848,
        -0.00810294,  0.00366683],
       ..., 
       [ 0.33449978,  0.02138252, -0.14221951, ..., -0.49634036,
        -0.47021103,  0.12775964],
       [-0.09816289, -0.34331754,  0.03284245, ..., -0.37847662,
         0.16994159,  0.44379511],
       [ 0.36513436,  0.40277815, -0.48408747, ..., -0.46500829,
         0.12536223,  0.04070676]], dtype=float32)



In [27]:

    
for v in tf.all_variables():
    print(v.name, v)









    



counter:0 <tensorflow.python.ops.variables.Variable object at 0x7f4f603ba208>
weights:0 <tensorflow.python.ops.variables.Variable object at 0x7f4f603ca278>
w2:0 <tensorflow.python.ops.variables.Variable object at 0x7f4f60369f60>
w_twice:0 <tensorflow.python.ops.variables.Variable object at 0x7f4f60369e10>

Feeds



In [28]:

    
input1 = tf.placeholder(tf.float32)
input2 = tf.placeholder(tf.float32)
output = tf.mul(input1, input2)
output.eval(feed_dict={input1: [9.], input2: [4.]})









    Out[28]:





array([ 36.], dtype=float32)



In [29]:

    
tfdot()









    Out[29]:



In [ ]: