Written by TRIPTI SINGH

Start with Tensorflow

Hello all! My previous blog posts were entirely focused on getting you comfortable with the idea of ML.

I decided for a change to bring to you something fun in ML. Highlight an important aspect of it i.e. DEEP LEARNING! We will get started with the tensorflow basics here.

TENSORFLOW is an API developed by google brain team.

An open source library for graphical based computation. It is used for: model building and deployment.

Tensor: is a generalization of vectors and matrices to potentially higher dimensions.In simple words it is a collection of numbers arranged in a particular shape.

                                                 N dimensional tensors

IT’S TWO BASIC OBJECTS OF COMPUTATION ARE : CONSTANTS AND VARIABLES

Constants : 

  • Cannot change
  • Cannot be trained 
  • Can have any dimension

Variables:

  • Values can be changed with computation.
  • Value is shared ,persistent and modifiable.
  • Fixed data type and shape.

Basic operations of tensorflow

                                             Tensorflow operation

Tf has a computational model that revolves around the use of graphs.

Tf graph has edges(tensors) and nodes(operations)

Add() operation to add two tensors

Matmul() operation to multiply two tensors.

Reduce_sum() operator sums over the dimensions  of a tensor (all or one)

Sum operation can be done with + operator as it is overloaded.

Addition can be  scalar,vector or matrix. Element wise addition takes place. Same shape required of 2 matrices is the pre-requisite condition.

A cheatsheet for addition

MULTIPLY OPERATION

TWO TYPES:

ELEMENT WISE AND MATRIX MULTIPLICATION

FOR ELEMENT WISE

  • USE MULTIPLY()
  • MUST HAVE SAME SHAPE
  • FOR MATRIX MULTIPLICATION

FOR MATRIX MULTIPLICATION

  • USE MATMUL(A,B) TO MULTIPLY A TO B
  • COLUMN NO. OF A SHOULD BE EQUAL TO ROW NO. OF B

Reduce_sum operator is used to perform summation over tensors , over all dimensions or one

Reduce_sum(A) all dimensions

Reduce_sum(A,i) over I dimension

Element-wise multiplication in TensorFlow is performed using two tensors with identical shapes. This is because the operation multiplies elements in corresponding positions in the two tensors. An example of an element-wise multiplication, denoted by the ⊙⊙ symbol, is shown below:

[1221]⊙[3215]=[3425]

ADVANCED OPERATION IN TF

  • GRADIENT ()- FINDS THE SLOPE OF A TENSOR AT A PARTICULAR POINT
  • RESHAPE()- RESFAPES THE TENSOR
  • RANDOM()-GENERATES TENSOR OUTTA RANDOMLY DRAWN VALUES

FINDING OPTIMUM -MINIMUM OR MAXIMUM ,BASICALLY NEEDED TO MINIMIZE THE LOSS FUNCTION

HOW TO DO IT??

VIA GRADIENT OPERATION

PASS VALUES TO GRADIENT FUNCTION UNTIL A POINT WHERE GRADIENT IS 0.

gradient increasing is minima and vice versa for maxima

Another gradient operation is

RESHAPING

Used extensively in image classification. Algorithms require to change matrices into vectors before classification.

Will be discussing image classification and some more cool content in the next blog. Be tuned!!

Please like, if you found this one helpful!

Leave a comment