Custom Vision | Azure - Datacloudy

 

In this blog we are going to see about Custom Vision in Azure, in crisp and clear manner. This is an important topic for those who are preparing for Azure AI-900 Certification Exam. So, Please follow all the points mentioned in this blog and this going to be super helpful for the certificate examination.

So what is Custom Vision,
As the name suggest, we can create models using our own images. There are 2 types and they are,

        i) Classification:

            In Classification the result may be of,
                       
                1) Multilabel : Multiple Tags per image

                2) Multiclass : Single tag per image

        ii) Object Detection :

              It returns Co-ordinates of object in the image.

Creating ML Models:

Before going in to this topic, we must know what is "Feature" and what is "Label". For that let us see below diagram,

            

From the flow we can understand very simple that, Feature and Label are nothing but "Input" and "Prediction" respectively. That is Model is created based on the input and the resultant Label is generated. Model is where the Machine Learning Algorithm comes into picture

    First  we need to choose the technique, and they are classified as 

• Supervised Learning 
• Unsupervised Learning
  

       Let us see one by one in brief manner,

    (*) Supervised Learning:

            It has Feature and Label. And it is of two types, they are Regression and Classification.

            

            i) Regression:

                   This is very very important topic as for as of certification concern. In Regression the Label is a numeric value with a range of possibilities.

Example : 
How much will it rain tomorrow?

                     In Certification Examination they give some scenario and ask for whether it is a Regression or Classification. For those question check for the output, if the output has numeric value with range of possibilities then we can choose Regression as answer. 

                     From the above example, The result of the question "How much will it rain tomorrow" may be "0.5 mm per hour". This is now clear that the result is in numeric value and it can be of any range of possibilities, thus it is regression  

            ii) Classification :

                    Here Label has limited set of Possibilities. That is, either 0 or 1.

Example:
Will it rain today?

                    In Certification Exam, check for the output for scenario based question. If the output has limited set of possibilities like true or false then blindly go with "Classification" as the answer. These are the thumb rule to tackle those questions. 

                    From the above example, The result of the question "Will it rain Today" may be 
"Yes" or "No". This is now clear that the result has limited set of Possibilities like yes or no therefore it is "Classification".

    (*) Unsupervised Learning:

                    It has no Label. And we need to know about Clustering here.

            i) Clustering: 

                    It Groups similar entities based on their features.

Example:
Divide customers into groups

Steps for Creating Machine Learning Models:

    Let us see the steps one by one,

    step 1: Obtain Data. This is the step where we gather the input data that act as Feature.

    step 2: Clean Data. In this step cleansing of data is done

    step 3:  Feature Engineering. It is the process where we  identify features and labels.

    step 4: Create a Model using the Dataset and the Machine Learning algorithm.

    step 5: Evaluate the accuracy of the model. It is like UAT that is, User Acceptance Test.

    step 6: Deploy the model for use. 

Thus in this blog, we saw about Custom Vision in Azure. We learnt about its types and also we saw the steps for creating the Machine Learning Model. Here the import point is we saw some of the Thumb rules that can be used to attend some of the questions in Azure AI-900 Certification Exam as well. Hope this information is helpful.

Thank You !!!