MLOps - Day 11

Today's learning :-

• As usual started with a quick revision of whatever studied last day, as a summery for limited data we can proceed with traditional(old) machine learning approach but if data is huge then we have to move on New ML or Deep Learning for fast & accurate results.
• In other word we can say that traditional ML usage sklearn(sci-kit-learn) in background and sklearn usage numpy in background and the way operations which numpy do is fine for limited data but when data is huge, it not sufficient, slow and less accurate.
• To solve this issue Google created a new data type Tensor, it is exactly similar like numpy and it is just a fancy name of array data type. Tensor is a data type which comes from TensorFlow module but Tensor is very much optimized because of graph. So finally we can say that Graph and Lazy execution is only the base of DL(Deep Learning)

 

• For below practical we will require Tensor module which is part of TensorFlow library, which is by default not part of anaconda Python distribution. So let's install using below command :-

C:\Windows\system32>conda env list
C:\Windows\system32>conda create --name mlops -y
C:\Windows\system32>activate mlops
(mlops) C:\Windows\system32>conda list
(mlops) C:\Windows\system32>conda install jupyter tensorflow pandas numpy matplotlib scikit-learn
It will take some time so wait till finish. Once finish you can start using it-
(mlops) C:\Windows\system32>jupyter notebook

• In TensorFlow version 2(Tv2) they support both lazy and eager execution but by default it execute with eager. 
• Lazy execution - In very simple words if I write code to perform an operation but don't run it now, run later when required, this kind of execution known as lazy execution. Eager execution example :- 

Code :-
import tensorflow as tf
tf.__version__
Output :-

'2.1.0' 

a = tf.constant(5)
b = tf.constant(6)
a,b
Output :-

(<tf.Tensor: shape=(), dtype=int32, numpy=5>,
 <tf.Tensor: shape=(), dtype=int32, numpy=6>)

#Eager execution
def myfun(a,b):
    c = a*b+1
    d = a*b*3
    print(c)
    print(d)
myfun(a,b)
Output :-

tf.Tensor(31, shape=(), dtype=int32)
tf.Tensor(90, shape=(), dtype=int32)  

•  But if we want can change it to lazy execution, how?

Code :-
import tensorflow as tf
a = tf.constant(5)
b = tf.constant(6)
#Lazy execution
@tf.function
def myfun(a,b):
    c = a*b+1
    d = a*b*3
    print(c)
    print(d)
myfun(a,b)
Output :-

Tensor("add:0", shape=(), dtype=int32)
Tensor("mul_2:0", shape=(), dtype=int32)