Monday, 12 March 2018

Supervised machine learning

In supervised machine learning we are learning to create models that combine inputs to produce useful predictions, these predictions are also produced even on unseen data. There is few key term in supervised machine learning.

supervised machine learning

Key term in supervised machine learning

  1. Labels
  2. Features
  3. Examples
    Labeled Examples
    Unlabeled Examples
  4. Models
  5. Classification
  6. Regression

Labels in supervised machine learning

A label is anything which we are going to predict. For example in mathematics if we are going to find or predict out the y value than here the label is y. Similarly the future price of gold is a label. In email filtering the label might be span or not span. In simple words the label is what? that we are trying to predict. 

Features in supervised machine learning

The features are the way that we represent our data. In other words a feature is an input variable. For example if we look into email example than feature might be the words of email, or, to or from addresses. The feature is any piece of information that we want to extract from data or in our case email to represent in supervised machine learning system.

In a simple machine learning project there are only single feature and in a complex machine learning project there are millions of feature in single project.

Examples in supervised machine learning

In machine learning everything revolves around the data, and an example is a particular instance of data. We can break examples into two different categories. The first one is labeled examples and the other is unlabeled examples.
  1. Labeled Examples
  2. Unlabeled Examples

Labeled Examples

In supervised machine learning we are going to use labeled examples to train our model. And in a labeled example we deal with both key terminologies features and label.
Labeled examples: {features, label} : {x, y}
Let’s take a look an example to understand labeled examples.
In labeled examples we obtain labels for data by asking humans to make judgments about unlabeled data for example
Does this image contains duck or hen
Whether the dot in x-ray is tumor or something else
In other words in labeled examples user actions are involved to obtain label for a given piece of unlabeled data. For example user performs some actions on email to mark as spam or not, and after obtaining labeled dataset we use this information to train our supervised machine learning model.

Unlabeled Examples

An unlabeled examples contains only feature. There is no explanation for each piece of unlabeled data it just contains the data, and nothing else. Basically unlabeled data consists of human created artifacts and we can obtain these from world.
unlabeled examples: {features, ?}: (x, ?)
Once we have trained our machine learning model with labeled examples, than we use that model to predict the label on unlabeled examples.


Let's understand supervised machine learning model with an example. Suppose we have asked to create a system that answers the question that an image has a dog or a cat. To give answer for that question we create a model or teach a model in supervised machine learning. Finally we have a model in that model we put the things that we are predicted. So in supervised machine learning a model define the relationship between features and label. Supervised Machine learning model life cycle has two phases.
  1. Training
  2. Inference


Training phase is the creating or learning phase of the model. In this phase we teach the model according to labeled examples. And enable the model to learn the relationship between features and labels.


In inference we apply the trained model to unlabeled examples and later used that trained model to predictions.


In supervised machine learning regression and classification both related to predictions. In regression we predict a value from continuous set of model or data on the other hand in classification we predict a value from discrete set of data. The example of regression is the value/price of house in a region, and the example of classification is "is this an image of dog, cat or something else".

In my next article we will discussed about linear regression.

No comments:

Post a Comment