Tensorflow basics

Machine learning might be frightening for beginners.

So let's learn something extremely simple so you could feel the ground.

Preface

The easiest way to play with tensorflow - is to use Google Colab notebook

Important !

Lets make 2 initial steps

import tensorflow as tf
tf.enable_eager_execution() # for tensorflow 1.x

Constants and Variables

In machine learning everything is represented in numbers. Images, music, just simple tables - all of it is represented in numbers. In mathematics there is a beautiful term for it - Tensor. So tensor - is just a set of numbers.

Sensor could be a single number - 1 or 2 or 100500
In TF you can define it as

d0 = tf.ones((1,))
d0.numpy()

# result
# array([1.], dtype=float32)

tf.ones - create a tensor where every number inside of is 1. Parameters which you pass inside of it - is dimentions.

d0 = tf.ones((5,))
d0.numpy()

# result
# array([1., 1., 1., 1., 1.], dtype=float32)

d0 = tf.ones((5,5))
d0.numpy()

# result
# array([[1., 1., 1., 1., 1.],
       [1., 1., 1., 1., 1.],
       [1., 1., 1., 1., 1.],
       [1., 1., 1., 1., 1.],
       [1., 1., 1., 1., 1.]], dtype=float32)

d0 = tf.ones((5,5,5))
d0.numpy()

# result
# array([[[1., 1., 1.],
        [1., 1., 1.],
        [1., 1., 1.],
        [1., 1., 1.],
        [1., 1., 1.]],

       [[1., 1., 1.],
        [1., 1., 1.],
        [1., 1., 1.],
        [1., 1., 1.],
        [1., 1., 1.]],

       [[1., 1., 1.],
        [1., 1., 1.],
        [1., 1., 1.],
        [1., 1., 1.],
        [1., 1., 1.]],

       [[1., 1., 1.],
        [1., 1., 1.],
        [1., 1., 1.],
        [1., 1., 1.],
        [1., 1., 1.]],

       [[1., 1., 1.],
        [1., 1., 1.],
        [1., 1., 1.],
        [1., 1., 1.],
        [1., 1., 1.]]], dtype=float32

Do display tensor as a table - use method numpy()
It's working only when Eager Execution mode is enabled (We did it in the very beginning). What is eager mode I'll explain in futher tutorials.

Constants

Constant - just takes any values and make it immutable. You can pass almost any value

from tensorflow import constant

credit_constant = constant(15)
# <tf.Tensor: id=18, shape=(), dtype=int32, numpy=1>

credit_constant = constant([13,4])
# <tf.Tensor: id=19, shape=(), dtype=int32, numpy=1>

So there are 2 methods available for constant dtype and shape

credit_constant.dtype
# tf.int64

credit_constant.shape
# TensorShape([Dimension(2)])

Variables

A1 = Variable([1, 2, 3, 4])
# <tf.Variable 'Variable:0' shape=(4,) dtype=int32, numpy=array([1, 2, 3, 4], dtype=int32)>
#cool, right?

#So then we can convert it to numpy array 
B1 = A1.numpy()

#array([1, 2, 3, 4], dtype=int32)

Funny fact

Whatever you set up - variable or constant - it's all tensors.

Follow me in twitter
@alexpolymath