Introduction

Hello world!

tf.constant: A tensor flow constant! Can be a string, number or a tensor. Once the value for a constant is set, it can never change!



In [1]:

    
import tensorflow as tf

# Create a tensorflow constant
hello = tf.constant("Hello World!")
# Print this variable as is
print(hello)









    



Tensor("Const:0", shape=(), dtype=string)

Oops! That is not what we wanted! This is because the variable hello hasn't been evaluated yet. Tensorflow needs a session to run the graph in!



In [2]:

    
# Create a new session
sess = tf.Session()
# Print the constant
print("Printing using Session.run()")
print(sess.run(hello))
# Also
print("Printing using eval() function")
print(hello.eval(session=sess))









    



Printing using Session.run()
b'Hello World!'
Printing using eval() function
b'Hello World!'

Tensorflow math operations!



In [3]:

    
# run addition and multiplication operations

a = tf.constant(25, tf.float32)
b = tf.constant(5, tf.float32)

with tf.Session() as sess:
    print("A = %f"%sess.run(a))
    print("B = %f"%sess.run(b))
    print("A + B = %f"%sess.run(a+b))
    print("A * B = %f"%sess.run(a*b))
    print("A / B = %f"%sess.run(a/b))
    print("A + B using tf add = %f"%sess.run(tf.add(a, b)))
    print("A * B using tf multiply = %f"%sess.run(tf.multiply(a, b)))









    



A = 25.000000
B = 5.000000
A + B = 30.000000
A * B = 125.000000
A / B = 5.000000
A + B using tf add = 30.000000
A * B using tf multiply = 125.000000

Tensorflow placeholder is a promise to provide a value later (supplied/fed at execution time). For placeholders, an optional argument shape can be used to make sure the input dimensions matches the required tensor dimensions.

If this is missing or None (default), then the placeholder can accept any shape.



In [4]:

    
# Run addition and multiplication with placeholders

c = tf.placeholder(tf.float32, shape=())
d = tf.placeholder(tf.float32, shape=())

sum = tf.add(c, d)
prod = tf.multiply(c, d)

with tf.Session() as sess:
    
    print("Operations by feeding values")
    print("C = %f"%sess.run(c, feed_dict={c: 4}))
    print("D = %f"%sess.run(d, feed_dict={d: 6}))
    print("Sum = %f"%sess.run(sum, feed_dict={c: 4, d: 6}))
    print("Prod = %f"%sess.run(prod, feed_dict={c: 4, d: 6}))









    



Operations by feeding values
C = 4.000000
D = 6.000000
Sum = 10.000000
Prod = 24.000000



In [5]:

    
# Matrix operations with placeholders
import numpy as np

mat1 = tf.placeholder(tf.float32, shape=(2,2))
mat2 = tf.placeholder(tf.float32, shape=(2,1))

matmul = tf.matmul(mat1, mat2)

with tf.Session() as sess:
    print("Matrix multiplication using python lists as feed dict values")
    print(sess.run(matmul, feed_dict={ mat1: [[1,2],[2,1]], mat2: [[1],[2]]}))
    print("Matrix multiplication using numpyarrays as feed dict values")
    print(sess.run(matmul, feed_dict={ mat1: np.array([[1,2],[2,1]]), mat2: np.array([[1],[2]])}))









    



Matrix multiplication using python lists as feed dict values
[[ 5.]
 [ 4.]]
Matrix multiplication using numpyarrays as feed dict values
[[ 5.]
 [ 4.]]