Tensorflow Basics

Remember to reference the video for full explanations, this is just a notebook for code reference.

You can import the library:



In [1]:

    
import tensorflow as tf

Simple Constants

Let's show how to create a simple constant with Tensorflow, which TF stores as a tensor object:



In [2]:

    
hello = tf.constant('Hello World')



In [3]:

    
type(hello)









    Out[3]:





tensorflow.python.framework.ops.Tensor



In [4]:

    
x = tf.constant(100)



In [5]:

    
type(x)









    Out[5]:





tensorflow.python.framework.ops.Tensor

Running Sessions

Now you can create a TensorFlow Session, which is a class for running TensorFlow operations.

A Session object encapsulates the environment in which Operation objects are executed, and Tensor objects are evaluated. For example:



In [10]:

    
sess = tf.Session()



In [15]:

    
sess.run(hello)









    Out[15]:





b'Hello World'



In [16]:

    
type(sess.run(hello))









    Out[16]:





bytes



In [17]:

    
sess.run(x)









    Out[17]:





100



In [14]:

    
type(sess.run(x))









    Out[14]:





numpy.int32

Operations

You can line up multiple Tensorflow operations in to be run during a session:



In [33]:

    
x = tf.constant(2)
y = tf.constant(3)



In [34]:

    
with tf.Session() as sess:
    print('Operations with Constants')
    print('Addition',sess.run(x+y))
    print('Subtraction',sess.run(x-y))
    print('Multiplication',sess.run(x*y))
    print('Division',sess.run(x/y))









    



Operations with Constants
Addition 5
Subtraction -1
Multiplication 6
Division 0.666666666667

Placeholder

You may not always have the constants right away, and you may be waiting for a constant to appear after a cycle of operations. tf.placeholder is a tool for this. It inserts a placeholder for a tensor that will be always fed.

Important: This tensor will produce an error if evaluated. Its value must be fed using the feed_dict optional argument to Session.run(), Tensor.eval(), or Operation.run(). For example, for a placeholder of a matrix of floating point numbers:

x = tf.placeholder(tf.float32, shape=(1024, 1024))

Here is an example for integer placeholders:



In [46]:

    
x = tf.placeholder(tf.int32)
y = tf.placeholder(tf.int32)



In [47]:

    
x









    Out[47]:





<tf.Tensor 'Placeholder_2:0' shape=<unknown> dtype=int16>



In [48]:

    
type(x)









    Out[48]:





tensorflow.python.framework.ops.Tensor

Defining Operations



In [30]:

    
add = tf.add(x,y)
sub = tf.sub(x,y)
mul = tf.mul(x,y)

Running operations with variable input:



In [31]:

    
d = {x:20,y:30}



In [32]:

    
with tf.Session() as sess:
    print('Operations with Constants')
    print('Addition',sess.run(add,feed_dict=d))
    print('Subtraction',sess.run(sub,feed_dict=d))
    print('Multiplication',sess.run(mul,feed_dict=d))









    



Operations with Constants
Addition 50
Subtraction -10
Multiplication 600

Now let's see an example of a more complex operation, using Matrix Multiplication. First we need to create the matrices:



In [69]:

    
import numpy as np
# Make sure to use floats here, int64 will cause an error.
a = np.array([[5.0,5.0]])
b = np.array([[2.0],[2.0]])



In [75]:

    
a









    Out[75]:





array([[ 5.,  5.]])



In [76]:

    
a.shape









    Out[76]:





(1, 2)



In [77]:

    
b









    Out[77]:





array([[ 2.],
       [ 2.]])



In [78]:

    
b.shape









    Out[78]:





(2, 1)



In [79]:

    
mat1 = tf.constant(a)



In [80]:

    
mat2 = tf.constant(b)

The matrix multiplication operation:



In [81]:

    
matrix_multi = tf.matmul(mat1,mat2)

Now run the session to perform the Operation:



In [82]:

    
with tf.Session() as sess:
    result = sess.run(matrix_multi)
    print(result)









    



[[ 20.]]